# AGO AGO is a complete customer OS for customer support, internal support, and pre-sales, packed with features that let teams craft a tailor made experience for every customer the moment they need help. AGO is a conversational layer that integrates directly inside our customers' product through our SDK, helping users at the right moment, before they even think about creating a ticket. When the AI cannot answer, the request is forwarded to the customer's ticketing tool (Zendesk, Intercom, Help Scout, etc.). At Mirakl, AGO cut ticket volume by more than 50%, and 96% of conversations are handled end to end by AI with no human intervention. Mirakl runs 25k conversations per month on AGO at 96% CSAT and 36% efficiency gains, and OpenAI and Mirakl have publicly promoted our partnership at https://openai.com/index/mirakl/. Other customers include Silae and Solocal, both large enterprises. AGO covers three agent surfaces in production: customer support agents that deflect and resolve tickets in product through the SDK, take actions in backoffice tools, and hand off to the customer's ticketing tool when needed; internal support agents for employees (HR, IT, ops) that answer questions and trigger workflows; pre-sales agents that guide prospects during their buying journey, qualify intent, and hand off to sales. Business teams and technical teams can use AGO autonomously and build on top of the platform on their own. AGO also has a forward deployed team that steps in whenever the customer needs help to design, integrate, or continuously improve the agent inside their product and their actual context. Pricing is 0.50€ per conversation with a 500€/month minimum; enterprise pricing is available for high-volume customers. The platform supports any language and runs 24/7. Every customer is paired with our consulting team, which sets the agents up with the latest best practices and stays involved as the agents evolve. The marketing site is bilingual (English and French). AGO's founders also created and run Generative AI Paris (https://www.generativeai.paris/), the largest French network of professional AI practitioners. ## Home AGO is the Customer Agent OS. AI agents that actually work, integrated into your stack. ### Analytics Fine grained control and visibility into your AI agents' performance. Real time dashboards and analytics to monitor agent effectiveness, customer satisfaction, and team performance. - Conversation volume: Daily trends and user activity. - Resolution rates: First contact and escalation metrics. - Knowledge usage: Top docs and content gaps. - Quality scoring: Human and AI evaluation metrics. ### Agent lifecycle Build and test AI agents, fast. 1. Create your agent. Spin up a new agent in minutes. 2. Plug a source of knowledge and APIs. Connect docs, ticketing, databases, APIs. 3. Test it in the agent lab. Run scenarios, replay real conversations. 4. Put it in production with access control. Decide who can talk to which agent. 5. Monitor the quality of every conversation. AGO evaluates and proactively looks for issues in conversations. 6. Improve the agent. Feed insights back. Compounding quality. ### Testimonials Featured by OpenAI: OpenAI published a story about our work with Mirakl. "On the client side, we are already seeing meaningful changes in usage. Some clients have fully integrated our digital agent into their daily routines. Whenever a question comes up or there is an internal discussion, their reflex is to turn to the agent." — Juliette Hendouze, Director Customer Services at Mirakl. "AGO permet à nos clients d'avoir une réponse à leurs questions tout de suite, permettant aux équipes support de se concentrer sur les problèmes compliqués" — Thibaut Desemberg, CEO, Tickee. ## Platform The AGO platform. AI agents that actually work. A complete platform to deploy AI agents that answer your customers, take actions in your systems, and continuously improve, with full visibility into what is happening. ### AI agents that take action in your backoffice Not just chatbots that answer questions. AI agents that actually do the work. - Process orders and refunds: Agents access your order management system to check status, process returns, and issue refunds autonomously. - Update customer records: Modify addresses, preferences, and subscription plans directly in your CRM without human intervention. - Trigger workflows: Escalate to the right team, create tickets, send notifications. Agents orchestrate your entire support workflow. ### Connected to all your documentation Your agents have access to everything they need to give accurate answers. - Any source, one platform: Connect Notion, Confluence, Google Drive, Zendesk, Intercom, your website, PDFs, and more. All your knowledge in one place, ready for AI. - Smart retrieval: Advanced semantic search finds the right information even when customers do not use the exact words. - Always up to date: Automatic sync keeps your knowledge base current. - Real time data: Query your databases and APIs in real time for order status, account info, inventory levels. ### Continuously improve response quality A complete toolkit to monitor, test, and enhance your AI agents over time. Conversation review, knowledge gap detection, A/B testing, prompt engineering, quality scoring, AI self improvement. ### Real time visibility into everything - Live dashboard for current conversations, resolution rates, response times, and CSAT. - Topic trends to spot emerging issues and product feedback. - Agent performance comparisons across channels. ### Configure agents for your use case Model selection (GPT, Claude, others), knowledge scoping, tool permissions, access control, prompt engineering, multi agent setup. ## DeepAgents AI that works for hours, not seconds. Most AI agents answer a question and move on. DeepAgents stay on the job. They run for minutes or hours, calling APIs, pulling data from multiple systems, and completing multi step tasks that used to require a human sitting at a screen all day. ### Not a chatbot. A persistent worker. A DeepAgent receives an objective, breaks it into steps, executes each one, and adapts when something goes wrong. - Runs until the job is done. A customer sends a billing dispute that involves checking three systems, verifying transaction history, and applying a partial refund. A DeepAgent handles the entire flow in one go, even if it takes 20 minutes. - Thinks in steps, not single answers. When a task requires checking eligibility, then looking up inventory, then processing a return, a DeepAgent sequences those steps and adjusts the plan if one of them returns unexpected data. ### Capabilities - API orchestration: Chain calls to your CRM, payment system, and warehouse API in sequence. - Cross system research: Pull information from your knowledge base, past tickets, and product docs. - Process execution: Run a full return workflow. - Live decision making: If an API call fails, the agent re evaluates its plan. - Task decomposition: Give it a high level goal and it figures out the steps. - Full context retention: Every piece of data the agent gathers stays in its memory. ### Use cases - Billing disputes: Pull transaction history, check the payment gateway, identify duplicate charges, process refunds. Three minutes instead of 25. Systems involved: Payment gateway, order management, CRM, email. - Returns and exchanges: Verify the return window, coordinate pickup, issue store credit or refund, send tracking. Systems: Order management, logistics API, payment, notifications. - Technical troubleshooting: Search the knowledge base, check the user's account configuration, test a fix, walk the user through the solution. Systems: Knowledge base, user accounts, sandbox environment, ticketing. - Escalation handoffs: Gather every piece of context first so the human picks up with zero ramp up time. Systems: Ticketing, CRM, conversation history, internal routing. ### How a DeepAgent runs a task 1. Understand and plan: Reads the objective, identifies systems, builds an execution plan with fallback paths. 2. Execute and adapt: Runs each step, checks the result, retries or reroutes when needed. 3. Deliver the outcome: A resolved ticket, a processed refund, a detailed report. Everything stays in memory. A DeepAgent remembers every API response, every piece of customer data, and every decision it made during the task. ### Results - Complex cases get resolved automatically. - Support teams focus on what matters. - Consistent quality at any volume. ## Smart ticketing AI powered ticket intelligence. Transform your support operations with intelligent ticket management that automatically detects patterns, groups similar issues, routes to the right teams, and resolves complex problems using deep AI agents, all while keeping stakeholders informed in real time. ### Automatically detect and group similar problems - Pattern recognition: Identifies similar issues across thousands of tickets, even when described differently. Groups related tickets to spot trending issues instantly. - Root cause analysis: Goes beyond surface symptoms to identify underlying issues. ### Deep agents for complex ticket resolution - Autonomous investigation: Analyze logs, check multiple systems, correlate data points, and build context before proposing solutions. Example: Customer reports intermittent service issues, agent checks error logs, correlates with system metrics, identifies pattern, tests hypothesis, proposes fix. - Multi step resolution: Handles tickets requiring multiple actions across different systems. Example: Account access issue, verify identity, check permissions, reset credentials, test access, update documentation, notify customer. ### Intelligent routing Skill based assignment, load balancing, performance optimization. Match tickets with team expertise; distribute by capacity and priority; learn from resolution patterns. ### Real time updates Milestone alerts, proactive updates, team collaboration. AI predicts when customers need updates and sends proactive communications. ### Natural language AI procedures Define complex resolution procedures in plain language that AI agents can understand and execute autonomously. Example: "When a customer reports a payment issue, first check their payment history, then verify the transaction status with the payment provider, and if the payment failed, provide alternative payment options." Procedures handle edge cases, learn from successful resolutions, suggest improvements, and adapt to changing business rules. ### Comprehensive analytics dashboard Trend analysis, resolution metrics, AI effectiveness, customer insights. Real time intelligence with predictive analytics, root cause reports, and performance optimization recommendations. ## AI ready documentation Transform scattered documentation into AI ready knowledge. Your documentation is everywhere: wikis, PDFs, Notion, Google Docs, ticketing systems. We turn this chaos into a clean, structured knowledge base that powers intelligent AI agents. ### The challenge Documentation chaos kills AI performance. - Scattered everywhere across Confluence, Notion, Google Drive, SharePoint, PDFs, emails. - Outdated and duplicated content. Multiple versions of the same document. - AI cannot use it. Poorly structured content leads to hallucinations and wrong answers. ### Our solution From chaos to AI powered excellence: complete audit, smart consolidation, AI optimization, platform integration, quality validation, ongoing sync. ### Process 1. Discovery and inventory. Map all your documentation sources. 2. Cleanup and deduplication. Remove outdated content, merge duplicates, identify the authoritative version. 3. AI optimized restructuring. Reformat content for optimal AI retrieval: consistent structure, clear headings, proper chunking, semantic organization. 4. Platform integration and testing. Import into AGO, validate accuracy, fine tune. ### Results Accurate AI responses, single source of truth, faster AI deployment, sustainable process. ### Why AGO Platform native optimization and end to end solution. We know exactly how AGO's AI agents retrieve and use information. From messy documentation to production AI agents. ## Why AGO Why leading companies choose AGO. We don't just provide AI tools. We partner with you to transform your entire customer operations. ### Why customers choose AGO Customers work with AGO for two reasons: exceptional service, and a platform they can actually build on. With other providers, they hit a wall the moment they tried to create agents that pull real time data from their own systems and sit natively inside their own product. Most of those tools were built for the support team, not for the end user. With AGO, you build AI agents that genuinely resolve user inquiries inside the product, so the volume your team has to handle drops dramatically and stops being a problem. Sales and product teams can also plug into AGO to capture valuable signals from prospects and customers. ### The customer support scaling crisis - Volume overwhelm: 30% yearly growth means exponentially more inquiries. - Global complexity: Multilingual support and regional knowledge sharing become impossible to manage consistently. - Quality versus speed: Maintaining response quality while keeping costs under control becomes a balancing act. ### Proven impact - 23k+ questions answered monthly. - 95% response accuracy rate. - Up to 70% ticket reduction. - 24/7 instant support. "Some clients have fully integrated our digital agent into their daily routines. Whenever a question comes up, their reflex is to turn to the agent." "Teams use the agent to draft emails, saving real time and allowing human effort to be refocused on higher value tasks." ### Built for enterprise integration Intelligent handoff, enterprise security, action oriented AI, true multilingual (24 languages), flexible data pipeline (Confluence, GitHub, Zendesk, PDFs), continuous learning. ### Where most AI solutions fall short - The "chatbot" limitation: Most are glorified FAQ bots. AGO's approach: DeepAgents handle multi step workflows. - Integration theater: Many vendors promise "seamless integration" but deliver rigid APIs. AGO's approach: True interoperability with your existing systems. - One size fits all mentality. AGO's approach: Custom built solutions that adapt to your unique business requirements. - Implementation limbo. AGO's approach: ROI within weeks through rapid deployment. The result: a new standard for AI customer service. While others deliver incremental improvements to broken processes, AGO fundamentally transforms how customer service operates. ### The AGO difference Rapid evolution, dual expertise, proven flexibility, measurable ROI. More than software, a strategic partner. ## Forward Deployed Engineers Forward deployed engineers. Custom solutions on the AGO platform. Our forward deployed engineers work directly with your team to build tailored AI solutions on top of our platform. ### The approach Engineering excellence, embedded in your team. - Embedded collaboration: Work side by side with engineers who integrate into your workflow. - Custom built: Every solution is designed specifically for your use case. - Rapid delivery: From concept to production quickly. ### Capabilities Custom integrations, advanced workflows, specialized AI agents, custom UI components, team training and support, scalable architecture. ### Process 1. Discovery and planning. 2. Rapid prototyping. 3. Custom development. 4. Deployment and knowledge transfer. ### Advantages Platform foundation, custom everything, deep expertise, faster time to value. ### Real world applications - Complex data pipeline integration. Result: 90% reduction in manual data updates. - Industry specific agent (healthcare compliance). Result: 100% compliance with faster response times. - Custom analytics dashboard. Result: Actionable insights driving strategic decisions. - Multi system workflow orchestration. Result: Hours saved per support case. ## AI Bootcamp 3 day AI bootcamp. From concept to production. Transform your real business use case into a production ready AI solution in just 3 days. ### Instant value - Real use case: Bring your actual business challenge. - Production ready: Walk away with a fully deployed solution. - Instant ROI: Start seeing returns from day one. ### The 3 day journey Day 1: Discovery and design. Deep dive into use case and requirements; data assessment and integration planning; AI solution architecture design; success metrics definition. Day 2: Build and integrate. Hands on AI model implementation; system integrations and API connections; custom workflow automation; testing with real data. Day 3: Deploy and improve. Production deployment; team training and handover; documentation and playbooks; go live and monitoring setup. ### What you walk away with Live AI solution, trained team, complete documentation, integration framework, performance metrics, 30 day support. ### Perfect bootcamp use cases - Custom care use cases with API and workflow integration. API and workflow integration; custom business logic; real time integration with external services. - AI internal tool powered by your documentation. Instant answers from all your documentation; process automation based on procedures; secure, compliant internal deployment. ### Prerequisites From your side: A specific use case with clear business value; access to relevant data and systems; 2 to 3 team members available for 3 days; decision making authority. From our side: Our AI platform; expert AI engineers; proven implementation methodology; post bootcamp support. ### Formats On site intensive, remote delivery, hybrid approach. ## Zendesk alternative Replace the Zendesk widget, keep your Zendesk. AGO is not here to rip out your help desk. AGO replaces the conversational layer (the widget and the chatbot) with autonomous AI agents that resolve most conversations end to end. Zendesk stays in place for ticketing and human agents, and AGO opens a ticket only when a human is truly needed. ### How it works AGO sits where your Zendesk widget used to be. It answers, takes actions in your IT systems, and resolves the conversation. When a human is needed, AGO creates a Zendesk ticket with full context attached. Your Zendesk instance, your routing, your SLAs, your reporting all keep working. ### Why support leaders replace the Zendesk widget - Your widget stops at FAQ deflection. Zendesk AI agents are good at intent routing and one shot replies but escalate as soon as the question needs context, a lookup, or a workflow. AGO replaces that widget with DeepAgents that read your knowledge, query your APIs, and resolve the conversation end to end. - Multilingual support is duplicated work. AGO answers natively in 24 languages from a single knowledge base. - Real integration with your IT systems. AGO connects deep into your IT stack so agents can check an order, refund a customer, update a subscription, or open a Jira ticket on their own. - You need a partner, not a portal. AGO gives you a forward deployed engineering team. ### Side by side | Capability | AGO | Zendesk widget | |---|---|---| | AI agents that resolve, not just deflect | DeepAgents handle multi step reasoning and trigger real workflows | Zendesk AI agents focus on intent routing and FAQ deflection | | Native multilingual coverage | 24 languages with automatic knowledge translation | Per language macros and content duplication | | Knowledge sources | Confluence, Notion, GitHub, PDFs, custom APIs, internal databases | Limited to Help Center articles and a few connectors | | Implementation | Forward deployed engineers ship with your team in days | Long onboarding, partner ecosystem for custom work | | Custom workflows from the widget | Agents take real actions across your stack from the conversation | Widget mostly opens a ticket and waits for a human | | Integration with your IT systems | Real connections to ERPs, CRMs, billing, internal APIs and databases | Marketplace apps focused on syncing data into the ticket | ### What you get with AGO DeepAgents that act, smart escalation to Zendesk, native multilingual, enterprise security, forward deployed engineers, continuous learning. ### Migration Week 1 to 2: Connect knowledge and IT systems. Plug AGO into your Help Center, Confluence, Notion, internal docs, ERPs, CRMs and product APIs. Week 3 to 4: Swap the widget. The Zendesk widget is replaced by AGO. Anything that needs a human is pushed back to Zendesk as a ticket with full context. Week 5 and beyond: Tune and scale. As resolution rates climb, fewer conversations turn into Zendesk tickets. ### FAQ Do I have to leave Zendesk to use AGO? No. AGO replaces the widget and the chatbot layer. Zendesk continues to be your ticketing system. How is AGO different from the Zendesk AI add ons? Zendesk's AI add ons are deflection layers focused on Help Center content. AGO is an autonomous agent platform that replaces the widget itself. When does AGO create a Zendesk ticket? Only when a human is truly needed. What does AGO cost compared to Zendesk AI add ons? AGO is priced on outcomes, mostly conversations resolved or workflows completed. How long does implementation take? Production value in weeks, not months. Can AGO handle complex B2B support? Yes. See the Mirakl case study. ## HubSpot Customer Agent alternative Replace the HubSpot Customer Agent, keep HubSpot. AGO becomes the conversational layer that answers with your knowledge and customer context, then calls your systems when it needs fresh data or an action. HubSpot remains the CRM and ticketing workspace. When a case needs a person, AGO creates a HubSpot ticket with the relevant context. Page: https://www.useago.com/integrations/hubspot-ai-support-agent/ ### Business and product Many personas. Give each persona its own knowledge, tools, instructions and tone while every audience remains connected to one HubSpot instance. Knowledge organization and quality. Organize knowledge across sources and document types. AGO adds scoring, gap detection, coverage tests and suggested changes instead of keeping private knowledge in one flat list. Testing before publication. Test conversations with several turns and mocked API responses before a change reaches customers. One CRM and ticketing tool. HubSpot remains native for CRM and tickets. AGO adds a separate conversational layer and sends escalations back with context. ### Technology and operations Custom customer experience. The AGO open source SDK lets teams build their own experience with React, Vue, Angular or TypeScript. Agents can navigate the application, update page state and call browser functions. Visibility. Teams can inspect tool calls, available reasoning traces, the system prompt and LLM timings. Actions. The prompt guides when the agent should act without requiring one exact trigger phrase. Long running work. Background agents can show the plan and progress while a task runs. Security. AGO can forward the user token to the customer backend, where existing permissions are applied to each tool call. ### Deployment and improvement Start with one concrete support case and connect the minimum knowledge and data required. Frequently needed context can be loaded at the start of the conversation. More specific data and actions can remain available through tools that the agent calls when needed. The agent lab supports conversations with several turns, mocked tools and agent duplication. In production, dashboards and automated review identify conversations that need attention. Teams can then improve the relevant prompt, documentation, API data or platform behavior. ## Pricing Simple, transparent pricing. Pay only for what you use. No hidden fees, no surprises. ### Pay per conversation (most popular) 0.50€ per conversation. Minimum 500€/month. Usage based pricing. Includes: unlimited AI agents; full integration with your systems; 24/7 availability in 24+ languages; analytics and reporting dashboard; dedicated onboarding support; GDPR compliant and secure. ### High volume Let's talk. Custom enterprise pricing. Processing more than 10,000 conversations per month? Contact us for custom enterprise pricing. Includes: volume discounts; dedicated success manager; custom SLA agreements; priority support and engineering; advanced security and compliance; forward deployed engineers. ### What makes us different No resolution concept. A conversation is counted when a customer starts an interaction. No complex resolution tracking or hidden metrics. Full platform included. Access the entire platform with unlimited seats. No per user fees, no hidden costs. ## Team ### Our mission Why we exist. We believe AI will drive real progress for everyone. We're simplifying the way people seek help to solve their problems. We do this by building AI agents that respond instantly and take the right actions to fix those problems. - Empower support teams: Tools that help customer support agents work more efficiently and effectively. - Enhance customer experience: Seamless, personalized customer experiences through intelligent technology. - Drive business growth: Scale customer operations without compromising on quality. ### Generative AI France community We cofounded the Generative AI France community to bring together Product and Tech professionals passionate about the transformative power of AI. Each month in Paris, we host a meetup focused on practical applications of generative AI, sharing experiences, and exploring new tools like LangChain. ### Founders Maxime THOONSEN, Co-founder. With extensive experience as CTO at Theodo, delivering custom, agile built solutions for demanding clients, Maxime knows how to translate complex technical innovations into practical, impactful results. His experience with open source AI, including contributions to LangChain and LLPhant which he created, makes him a pragmatic builder focused on what actually drives value and impact for AGO. Damien MOUROT, Co-founder. Damien leverages over 10 years of AI expertise to ensure AGO delivers powerful, scalable solutions tailored to your business. Having previously scaled AI products serving millions of users at Leboncoin, Damien understands how to align advanced technology with your strategic goals. His track record, including ranking in the global top 500 competitors on Kaggle, guarantees AGO brings world class AI capabilities directly into your hands. ### Join us We're always looking for talented individuals passionate about revolutionizing customer operations. Right now, we're interested in fullstack developers who want to work on cutting edge AI solutions. ## Case studies Customer success stories. Discover how organizations are transforming their customer operations with AGO. ### Featured: Mirakl Enterprise marketplace SaaS platform. Mirakl, a leading enterprise marketplace provider with $160M in annual recurring revenue, transformed their customer operations with AGO's AI powered solution. 23k conversations per month, 96% CSAT, 36% efficiency gain. ### Mirakl case study (full) How Mirakl Transformed Customer Support with AGO. "AGO has transformed our customer support operations. Customers now get instant, accurate answers, freeing up our team to focus on complex issues that truly need human insight." — Juliette Hendouze, Director Customer Services at Mirakl. Metrics: 23k questions answered with AI each month. 96.2% CSAT. 36% efficiency gain. Company overview: Mirakl is a leading enterprise marketplace SaaS platform provider with $160 million in annual recurring revenue and over 750 employees worldwide. Their platform powers major online marketplaces for brands like Decathlon and Leroy Merlin, offering marketplace and dropshipping solutions along with complementary tools for retail media and payment services. Challenge: With a 30% yearly growth, Mirakl faces challenges in scaling their customer support operations. They have a growing volume of customer inquiries and a need for multilingual knowledge sharing across regions while trying to maintain consistent response quality without costs spiraling out of control. Solution: Mirakl partnered with AGO to implement an AI-powered customer operations hub, branded internally as "One Help." This comprehensive solution provided an intelligent self-service platform, multilingual support with proper context, knowledge integration from documentation and support tickets, and analytics to identify knowledge gaps. The implementation process included knowledge corpus analysis, integration with Zendesk, GenAI team upskilling and continuous improvement on the documentation. Customer quotes: "I love One Help, I'm using it every time I have a question and systematically before opening a ticket. The answers are precise and very efficient. Congratulations to the whole team." — Mirakl Customer, Marketplace Operator. "We use One Help as our first line of support. It efficiently answers most of our initial questions and helps us identify and address any gaps or inconsistencies in the documentation, greatly improving our overall user experience." — Alice Henry, Manager, Technical Support at Mirakl. "I love it. I think it's the best thing you have released so far along with the search bar." — Marketplace Operator, Mirakl Client. "Je suis avec showroom et ils adorent le One Help! Ils m'ont dit que c'était super intuitif et que les réponses étaient tops." — Candide, Marketplace Consultant at Mirakl. "I just wanted to say that Mirakl One Help has been pretty helpful for me over the past few days! I got clear answers with a good level of detail, and in the languages I needed for my emails (Italian & Spanish)." — Support Team Member, Mirakl. ### Temana Import case study (full) Temana Import: an AI assistant that sells equipment 24/7 in French Polynesia. "We wanted our customers to get a quote at 3am just like at 10am. Today that's a reality, and the sales team focuses on closing deals." — Nuiata Sachet, Managing Partner, Temana Import. Metrics: 24/7 availability day and night, even on weekends. 227+ product sheets accessible in seconds. 3 languages: French, Tahitian, English. Company overview: Temana Import has been the go to professional equipment supplier in Tahiti for over 20 years. Sales, rental, maintenance: the company covers all needs for construction machinery, trucks, agricultural and forestry equipment across French Polynesia. Three complementary entities (Temana Import, Temana Location, Tahiti Flexibles Hydrauliques) serving construction, agriculture and industry. Challenge: Messenger is the primary contact channel in French Polynesia. Every day, Temana Import receives dozens of inquiries: price of a mini excavator, truck availability, specs on a crusher. The sales team was spending considerable time manually answering these questions across a catalog of 200+ products. In the evening, at night and on weekends, messages went unanswered until the next business day. Solution: AGO deployed an AI assistant on Facebook Messenger and the Temana Import website. The assistant queries the full catalog of 227+ products in real time to provide technical specs and exact pricing. It speaks French, Tahitian and English to adapt to each customer. When a prospect is qualified, it collects their contact details and routes the request to the sales team through an automated workflow. Result: Since launch, Temana Import customers get a response in seconds, no matter the time of day. The sales team receives qualified leads with contact details and the exact customer need, ready to be processed as soon as they open. Evening and weekend inquiries no longer go unanswered. User flow: Website chat (customer asks a question), AI agent (instant qualified answer), CRM (lead and context synced), Sales callback (rep closes the deal), Transaction (order confirmed). ## Integrations Connect AGO to your existing tools. AGO integrates with your knowledge bases, ticketing systems, and developer tools. Deploy AI powered customer support that works within your existing workflow. Categories: Knowledge sources (documentation, wikis, databases), Ticketing systems (helpdesks for ticket resolution), Developer tools (APIs and webhooks for any system). Available integration pages: - Airtable: AI support agent on top of your Airtable bases. - Confluence: AI support agent reading your Confluence spaces. - Crisp: AI support agent inside your Crisp inbox. - Custom API: Connect AGO to any internal API or system. - Help Scout: AI support agent inside Help Scout mailboxes. - HubSpot: Replace the HubSpot Customer Agent while keeping HubSpot for CRM, tickets and human support. AGO adds customer context, secure actions, many personas, testing, custom experiences and continuous improvement. - Intercom: AI support agent in Intercom conversations. - Jira: AI support agent that reads and acts on Jira issues. - MadCap Flare: AI support agent powered by MadCap Flare documentation. - n8n: AI support agent triggered through n8n workflows. - Notion: AI support agent reading your Notion workspaces. - Slack: AI support agent inside Slack channels. - Smart Tribune: AI support agent on top of Smart Tribune knowledge. - Trello: AI support agent integrated with Trello boards. - Zapier: AI support agent connected through Zapier automations. - Zendesk: AI support agent integrated with Zendesk tickets. Seamless escalation: When AI cannot fully resolve an issue, AGO creates rich tickets with complete context for your support team. Multi system support across Zendesk, HelpScout, Intercom, HubSpot. Rich context transfer including complete conversation transcript, customer metadata, knowledge articles referenced, and priority and category auto assignment. Two way messaging, custom fields, smart routing. ## Blog ### AGO at TECH FOR RETAIL 2025 Date: 2025-11-06. AGO will be attending Tech for Retail on November 24 and 25 at Paris Expo Porte de Versailles. Come and discover our AI agents capable of acting directly within your business tools and automating up to 96% of your customer requests. Tech for Retail: The Event Defining the Future of Retail Through Technology. Tech for Retail has become the must attend event for retail and tech professionals. It offers a complete and relevant overview of the latest digital tools and sustainable technological innovations for the retail industry. From Supply Chain to Customer Loyalty, technology is transforming retail at every level, every single day. This 5th edition (November 24 and 25, 2025 at Paris Expo Porte de Versailles) will highlight concrete advances in generative AI and automation in retail. In this context, AGO will showcase its latest innovations in AI for customer support. Why AGO is Participating in Tech for Retail. Because customer service challenges in retail have never been more strategic. Brands must handle an ever growing volume of requests across multiple channels, while ensuring fast and personalized responses. At AGO, we believe artificial intelligence shouldn't just reply, it should act. Our AI agents are built to perform real actions within your business tools, following your rules, safely and reliably. The result: up to 96% of customer requests automated, enhanced operational efficiency, and support teams refocused on high value interactions. Through tailored integration and full support (scoping, training, monitoring), AGO already helps companies like Mirakl transform their customer experience and reduce operational costs by over 36% even amid strong growth. What You'll Discover at Our Booth V18. Live demos of AI agents handling real world use cases (product returns, order tracking and updates, access resets, and more). Real client success stories, such as Mirakl, where AGO already automates 96% of customer support. A presentation of the AGO platform, designed to integrate with any information system, even complex ones. Personalized discussions with AGO's AI experts to explore automation opportunities for your business. Get Your Free Badge for Tech for Retail. Don't miss the opportunity to meet our team and discuss your customer service digitalization challenges. Meet us at booth V18, November 24 and 25, 2025, at Paris Expo Porte de Versailles. Two days to discover, get inspired, and rethink customer support in the age of AI. ### Offer 24/7 multilingual phone support with GPT Realtime and AGO Date: 2025-08-28. The future of customer support is no longer just text based. With the release of gpt-realtime, OpenAI has made production ready voice AI agents a reality. At AGO, we've taken it one step further: combining this technology with our own voice support feature, delivering 24/7 AI powered phone support that matches the quality of text based service. Why Voice AI Is Hard. Voice AI isn't just about generating speech that sounds human. The real challenges lie deeper. Interruptions: Human conversations are messy. People interrupt, change topics, or backtrack mid sentence. Most models fail at handling these shifts without breaking the flow. Language understanding: Models are weaker in French and other non English languages. Misunderstanding loops: When they misinterpret something, the wrong text remains in memory, and the agent continues down the wrong path. Hallucinations: Voice models sometimes invent information, which in a live support call can be disastrous. These pain points are why most companies have struggled to deploy voice AI in real world customer support. How AGO Solves It. At AGO, we built a voice first support system that eliminates hallucinations and misunderstanding loops. API verified answers: When the model doesn't know something, the AI Voice system calls our AI agents to retrieve the correct information. This prevents the model from guessing and ensures consistent reliability. Human level interruptions: Our infrastructure is optimized for overlapping speech, so the conversation feels natural, like talking to a skilled human agent. Context engineering: We've developed techniques to maintain context even when the conversation jumps around. This means the agent can handle complex, multi turn dialogues without losing track. The result: voice agents that are both reliable and trustworthy. GPT-Realtime Changes the Game. The launch of gpt-realtime introduces new capabilities that make our system even stronger. Asynchronous function calling: The agent keeps the dialogue flowing while waiting for results. Smarter comprehension: The new model is better at following complex instructions, handling alphanumerics, and even switching languages mid sentence. MCP integration: This allows everyone to integrate faster with external systems that publish their MCP servers. Also, with voices like Cedar and Marin, conversations feel warmer, more engaging, and ready for production. Why It Matters. Customer expectations have shifted. They no longer want to wait until office hours or navigate rigid IVR menus. They expect immediate, high quality support, anytime, anywhere. With AGO and gpt-realtime, a customer can call at 2 AM and speak with an AI agent that answers as reliably as text based chat. Businesses can scale voice support without scaling headcount. Conversations flow naturally, without the frustration of robotic scripts or broken context. This is AI powered customer service at human scale. What's Next. At AGO, we believe voice is one of the final frontiers of customer experience. Text bots have matured, but phone support is still where loyalty is won or lost. Thanks to gpt-realtime and our hallucination resistant AI agents, businesses can now deliver always on, human quality voice support. ### Create Agents and AI Tools with AGO's MCP server Date: 2025-11-16. Building powerful AI systems requires two key components: tools (integrations with APIs, search capabilities, and custom actions) and agents (the AI entities that use those tools to accomplish tasks). The Problem: AI Agent Management Doesn't Scale. As AI systems grow, you often need to create multiple agents and tools. In a typical agent deployment, you have HTTP tools for different API endpoints, various knowledge bases for the business context, and custom integrations for specific business systems. AGO already provides two ways to create these. The web UI is great for non developers who want a visual interface to click through forms. REST APIs are for developers who want to automate deployments with custom scripts. Managing these manually through a web interface can be tedious and error prone, especially as the number of agents and tools increases. Now, with MCP (Model Context Protocol) support, creating agents inside AGO becomes dramatically simpler. Just connect Claude Code or any MCP client to your AGO instance, and you can create, update, and explore both tools and agents using natural language, no scripts required. Real World Use Cases. One of the most powerful use cases is integrating AGO directly into Claude Code or any MCP compatible personal AI assistant. Add AGO's MCP server to Claude Code with: claude mcp add transport http ago http://your-server.com/api/mcp header "X-API-KEY: YOUR_API_KEY". It makes it way easier to build any agent or tool. Compare what we need to do when we Build a Returns Agent. Traditional approach: 1. Log into AGO dashboard. 2. Navigate to Tools, Create HTTP Tool. 3. Configure order lookup endpoint. 4. Navigate to Tools, Create another HTTP Tool. 5. Configure return processing endpoint. 6. Navigate to Tools, Create another HTTP Tool. 7. Configure refund endpoint. 8. Navigate to Agents, Create New Agent. 9. Select the three tools from dropdowns. 10. Write the agent prompt. 11. Configure knowledge sources. 12. Test and debug. With MCP in Claude Code you just need a prompt: Create a returns processing agent that can look up orders, process returns, issue refunds. The agent should verify order eligibility before processing returns and explain the process to customers. Claude creates three HTTP tools with the correct configurations, creates an agent with an appropriate prompt, assigns all three tools to the agent, and suggests test cases. Result: from 12 manual steps to a single conversation. Other examples: Create tools ("Create an HTTP tool that connects to our payment API"). Update configurations when systems change ("The user search endpoint changed its schema. Update the search_users tool to match the new spec."). Explore your infrastructure ("What agents do we have configured and what tools do they have access to?"). This is AI building AI infrastructure. You're not manually configuring each component, you're having a conversation about business requirements, and Claude translates that into working AI systems. Self Improving Systems. This opens up many possibilities: agents that extend themselves; infrastructure as conversation; rapid prototyping; version control for AI. What's Supported in this Release. The MCP server endpoint at /api/mcp provides capabilities for both tools and agents. Managing Tools (full CRUD): list_tools, get_tool, create_tool, update_tool, delete_tool. Managing Agents (full CRUD): list_agents, get_agent, create_agent, update_agent, delete_agent. What's Next. The MCP API for tools and agents is just the beginning. We're working on advanced analytics on agent and tool performance, and multi agent orchestration to define complex workflows involving multiple specialized agents. ### GenAI Days Paris 2026: Recap of Every Talk Date: 2026-04-17. I really loved this 4th edition of GenAI Days on April 16, 2026 in Paris. What made the day special for me was how genuinely insightful every single talk was. No filler, no vendor pitches dressed up as content, just builders sharing what actually works when you ship agents in production. I left with a full notebook and a lot to think about, and that is exactly why I wanted to write this recap. Nikolay Martynov (Bankast): The GenAI News Rundown. Nikolay gave his now classic tour of the latest releases, and this year he raised the bar with a new compression challenge: 60 slides in 15 minutes. One of the clearest signals through the firehose was that Anthropic is in very strong form at the start of the year, with a dense wave of releases outpacing the rest of the field. He covered Anthropic's Mythos model, the famous Claude Code leak where someone unpacked the .dev bundle and turned it into a masterclass on agent prompting, the controversy around Cursor 3.0's Composer 2 allegedly trained on Kimi K2, and Gemma 4 with its turbo quantization that divides RAM usage by six and runs comfortably on a single 4090. He also touched on GLM 5.1 and the caveman language token trick that saves context by compressing prompts into an ultra terse format. His takeaway was that the pace of frontier labs is still accelerating, and what worked six months ago is already an outdated pattern. Quote: "If you are not reading model release notes every week, you are already shipping last year's product." Key learnings: Open weights models are now close enough to frontier models that the choice is mostly about cost and latency. Leaked internals from production agents are the best free training material available. Token compression tricks like caveman language buy real context room when you need it. Anaïs Guesny and Antoine Habert: SaaS Is Dead, Long Live Lovable. Anaïs and Antoine told the story of building Quiverplace, a full marketplace shipped 100% with Lovable. They shared the raw numbers: roughly 800k lines of code generated and an estimated 27,000 hours of engineering time saved. They also told the very funny story of the day Lovable and Claude Code blamed each other for a bad commit, which forced them to give each AI its own scoped cloud.md and system prompt so they would stop overwriting each other's work. They used a Get Stuff Done style phasing approach to keep scope under control, and insisted that the real skill is no longer coding but writing tight specs and reviewing output with taste. Antoine who is a skilled architect was surprised by the quality of the code written by Lovable, he had less refacto work to do that he expected. Quote: "The hard part is not generating the code anymore, it is knowing exactly what you want before you ask for it." Key learnings: Product builders without a traditional engineering background can now ship production software. Multiple coding agents working on the same repo need clear scoping. The bottleneck has moved from writing code to writing specs and reviewing output. Léo Arsenin (Cloudflare): Secure Agentic AI at the Edge. Léo presented Cloudflare's vision of agentic AI running as close to the user as possible. He introduced Code Mode, a technique that compresses 2500 API endpoints into roughly 1000 tokens of context by letting the model write code that calls a small set of search and execute tools instead of binding every endpoint as a separate tool. The trick is radical in its simplicity: instead of exposing hundreds of tools, you expose only two, a search tool to find the right API and an exec tool to run the code the model just wrote against it. The model does the composition in code, which is what LLMs are already great at, and the token budget stays small no matter how big the underlying API surface is. He showed how npm install agent fits into a world of 330 data centers and per millisecond CPU billing, and made the case that running LLMs at the edge changes both the economics and the security model of agents. Latency drops, data residency becomes easier, and the surface area for prompt injection can be contained. Quote: "Tool calling does not scale. Code mode does." Key learnings: When you have more than a few dozen tools, let the model write code against a small set of primitives instead. Edge deployment is becoming a real option for agent workloads. Security posture for agents needs to be designed at the infrastructure layer. Raouf Chebri (Replit): Building Cooperative AI. Raouf built on Karpathy's line that the hottest new programming language is English. He laid out three pillars for cooperative AI (raw intelligence, verification, and context) and four traits of the modern builder: knowing what you want, taste, composition, and parallelism. He closed with a preview of Replit Agent 4 and a live demo of multiple agents collaborating in the same workspace without stepping on each other. Quote: "The best builders of the next decade will not be the best coders. They will be the best at knowing what to ask for." Key learnings: Verification is the bottleneck of cooperative AI, not intelligence. Parallelism is a learned skill. English is not replacing code, but it is replacing a lot of the code you used to write by hand. Ferdinand Legros (Heal.dev): QA Autopilot. Ferdinand demoed QA Autopilot, a testing agent built on top of Playwright locators rather than raw DOM exploration. He placed his product on a spectrum from pure agent to fully structured code and argued that the sweet spot for QA is closer to structured code than most people think. The headline numbers were sharp: about 2x faster and 2x cheaper than a stack combining Claude Code with Playwright, because the locator layer removes a huge amount of redundant exploration. He also shared that around 20% of end to end tests break every week in a fast moving product, which is why maintenance, not creation, is where agents create the most value. Quote: "The cost of tests is not writing them. It is fixing them every Monday morning." Key learnings: QA is one of the first engineering jobs that can be almost fully delegated to agents. Locators are a much better abstraction than raw DOM. Pick the right point on the agent to structured code spectrum. Maxime Thoonsen (AGO): The Rise of In App Agents. I mapped out three models of the agentic UX: MCP apps where SaaS gets absorbed into a chat, and in-app agents that live inside the product itself and take real actions on behalf of the user. It's an ingenuous way of solving the tradeoff of adding new features without bloating the app for nothing as the AI can pull the not so often used UI only when needed. I shared the architecture we use at AGO, built around JavaScript function calling tightly scoped to the user's current context. Quote: "In-app agent are a pragmatic way of bringing an agentic experience to your users." Key learnings: In-app agentic chat are the last trend to help user with agentic behavior. JavaScript function calling is a simple and underrated way to ship agents today. Support and onboarding flows are the first place to look for in app agent wins. Quentin Pleplé (Theodo): Agents That Code While You Sleep. The Health Hero migration, originally scoped for 70 weeks was delivered in 21 with an agent fleet. Quentin showed Theodo's modernization factory, where specialized agents run overnight in the cloud and pick up tickets while humans sleep. He framed the core problem with a simple formula: speed plus variance equals chaos, and explained how the factory controls variance through pipelines, guardrails, and review loops. Agents are triggered from GitHub Actions into sandboxed VMs that run during the night, applying a suite of skills to ensure quality. He insisted on doing kaizen everytime the coding agent produce a result with a default. They have a dedicated kaizen agent continuously improving the playbooks the other agents use. The migration plans themselves are 150 lines long max. To keep them short, they are packed with references to legacy code, which is what makes the agents reliable. Quote: "You do not need smarter agents. You need a smarter factory around them." Key learnings: Background agents shine on well defined, repetitive, high variance tasks like migrations. The review loop plus kaizen is where quality is won or lost. Alexandre Castera and Théo Dullin (Adaptive ML): Fine Tuning That Actually Works. Alexandre and Théo made the case for reinforcement learning as the real unlock of fine tuning. They opened with the classic GPT-3 graph showing that a prompted 175B model roughly matches a fine tuned 6B model, and used a simple analogy to explain RL versus supervised fine tuning: SFT is rote learning, RL is a student with a teacher who grades every answer. They shared an airline chatbot case study and a benchmark where Gemma 3 12B tuned with their method matches Sonnet 4.6 on a specific production task at roughly 4x lower cost and better response time. They also shared that their customers now push about 1T tokens per year through tuned open models. Quote: "Prompting gets you to a demo. RL gets you to production economics." Key learnings: RL fine tuning is moving out of frontier labs and into applied teams. A small tuned open model can beat a large frontier model on a narrow task, cheaply. Think of fine tuning as a cost optimization strategy. Charles Sonigo (Alpic): AI Native, MCP, and ChatGPT Apps. Charles walked through what it takes to build MCP apps that plug directly into ChatGPT, now that more than 300 such apps are live. He used two case studies: Excalidraw for real time collaboration inside ChatGPT, and Recommerce for a commerce flow where teenagers send a picture of a phone and get a diagnosis and a buyback offer in seconds. He detailed the UX surfaces available (picture in picture, full screen, inline cards) and the ranking signals that decide which app gets called, including negative prompts that let you tell ChatGPT when not to route to you. His main lesson was that distribution inside ChatGPT is a genuinely new channel, and the teams who land there first will define the category. Charles also invited the CEO of Recommerce on stage, whose main insight was simple and strategic: young users spend a huge share of their screen time inside ChatGPT, so shipping a ChatGPT app is one of the most efficient ways to reach and convert this audience, well ahead of classic e commerce channels. Quote: "ChatGPT just became the new App Store. Most companies have not noticed yet." Key learnings: MCP is turning traditional backends into AI native surfaces. Distribution inside ChatGPT is a new acquisition channel. Picking the right UX surface matters a lot for engagement. Katia Gil Guzman (OpenAI): Codex Inside OpenAI. Katia gave a rare look at how OpenAI itself uses Codex internally. Few hours after her talk, OpenAI published "Codex for almost everything", which spelled out most of what Katia had previewed on stage. What you can now do with the Codex App: Computer use on macOS, where Codex can observe the screen, click, and type with its own cursor; built in browser where you can annotate web pages directly; image generation with gpt-image-1.5; full SDLC support including PR review comment handling on GitHub, multiple terminal tabs, remote devboxes over SSH (alpha), rich file previews, and a summary pane that tracks the agent's plans, sources, and artifacts; 90+ new plugins that mix skills, app integrations, and MCP servers; long running work where automations can resume existing threads, schedule their own follow up tasks, and wake up over days or weeks; memory (preview); proactive task suggestions. She shared that 100% of PRs inside OpenAI are reviewed by Codex before a human ever sees them. The message was clear: agent workflows are not a future bet at OpenAI, they are already how the company ships software every day. Quote: "Codex is not a tool we ship. It is how we ship." Key learnings: Codex has become the central operating surface at OpenAI. The internal config stack (agents.md, skills, execution plans, plugins) is what makes Codex reliable at scale. With 100% of PRs reviewed by Codex first, the human loop has shifted from writing and reviewing code to directing agents. Alexis d'Eudeville (Lemlist): A PM Shipping to Prod. The most important insight from Alexis was not that a PM can now ship code. It was how PMs and devs redefine their respective jobs together, in real conditions, at a 15 product / 45 engineer company with 20k customers and 50M ARR. Alexis framed the new setup as a trio (dev, designer, PM) where the frontiers move but the roles do not merge. The road to get there was not smooth. First phase: early vibe coding went badly. PMs started a little internal contest of who ships the most to prod with their name on it, which the devs hated. Lemlist had to pull back, rethink the workflow, and set real rules. Second phase: when Claude Code was rolled out to the dev team, nothing shipped for a month. The PMs were looking at each other wondering what was happening. The devs were rewiring their habits, and it was painful. Today the team has stabilized around PMs with GitHub access, on demand environments, spec files generated directly from the codebase as .md, and a real focus on review as the new bottleneck. Quote: "Devs are scared. PMs are too. And that is normal, because everyone is redefining their job at the same time." Key learnings: The real story is PMs and devs renegotiating the contract between their roles. Devs do not escape the J curve. Review is the new bottleneck. Didier Lellouche (LCL): Background Agents in a Bank. Didier gave one of the most grounded talks of the day, interviewed on stage by Julie Crauet, Head of Product at Gojob, who led the discussion and pushed him on how LCL runs background AI agents inside a major French bank. LCL processes around 22 million emails per year, about 70% of employees already use an internal AI tool, and the succession dossiers case study saw error rates drop from 20% to 2% after introducing agent assistance. He walked through the governance side as well: a public commitment to no layoffs tied to AI, the CSE consultation required by French law, the distinction between mobilizable and non mobilizable ROI, and the 8 day security patch rule that shapes every vendor choice. He closed by pointing at Revolut as the non bank competitor that forces traditional banks to move faster. Quote: "One month of dev. Nine months to pass security, compliance, and works council approvals." Key learnings: Background agents are already producing measurable ROI in regulated industries. Governance is a first class design constraint. Non bank competitors are the real forcing function. Matthieu Dinot (Mistral): Devstral 2 and Mistral Vibe. Matthieu closed the day with a deep dive into Devstral 2 and Mistral Vibe. He explained how RL training pipelines work in practice on a Kubernetes based infrastructure, and shared a series of memorable reward hacking anecdotes, including a model that learned to curl GitHub directly to find the real PR patch instead of solving the task, and a submit tool whose distribution at training time did not match inference and quietly wrecked performance. He showed strong SWE bench verified numbers and made a clear case for open weights coding models: competitive quality, full control, and the ability to tune them for your own stack. Quote: "Every reward hack is a lesson in what you actually asked for versus what you thought you asked for." Key learnings: Open weights coding models are genuinely competitive on real world SWE tasks. Reward hacking is feedback on your reward design. Training and inference distribution mismatches are an underrated source of silent regressions. Closing. The strongest signal from this edition is that agents are no longer a lab topic. They are shipping in banks, in marketplaces built by two founders, in QA pipelines, in ChatGPT itself, and inside the IDEs of every speaker on stage. ### Imagine Summit 2025: The Developer in the Age of Generative AI Date: 2025-12-01. I'm excited to announce that I'll be speaking at Imagine Summit on December 4th in Rennes, France. Thanks to Le Poool x La French Tech Rennes St-Malo for the invitation. The Developer Role is Transforming. I'll be discussing a topic that's accelerating rapidly: the transformation of the developer role in the age of generative AI. We're moving from early coding agents to agent clusters, capable of coordinating and working in parallel on complex tasks. Two Pillars for the Future Developer. In my view, the future of the profession is built around two pillars. First, be the architect of the solution: make the right technology choices, understand where the trade offs lie, manage the flows and arbitrations. The developer becomes the one who designs the overall architecture and makes strategic decisions. Second, design the software factory: create an environment where agents write, test, document, and fix code with near autonomy. The developer orchestrates an AI powered software production pipeline. A Concrete Example: Automating Sonar Fixes. I developed a script that fetches Sonar issues and generates small Markdown files. Claude Code can then read them, understand them, and propose fixes autonomously. It's a small piece of the software factory, operational right now. I'm making it open source so anyone can have a look and adapt it. The AI Adoption Ladder for Developers. Earlier this year, Steve Yegge described the stages of AI adoption by developers. We're clearly moving up the ladder, faster than I expected back in March. 1. Code Completion (intelligent autocompletion). 2. Chat Coding (coding with a conversational assistant). 3. Vibe Coding (coding by describing intent). 4. Coding Agents (agents that code autonomously). 5. Agent Clusters (where we're arriving: agents that coordinate). 6. Agent Fleets (the logical next step: fleets of specialized agents). Panel Discussion: From Code to System Orchestration. During the panel, I'll be exchanging ideas with Jacques Le Mancq (Broadpeak) and Kim Bourget (Groupe SNCF) to compare perspectives from tech, industry, and public innovation around this profound shift. We're moving from developers who write code to developers who design and orchestrate software production systems. ### Maxime Thoonsen on Comptoir IA: Building the Future of Customer Operations Date: 2025-07-30. We're excited to share that our CTO and Co founder, Maxime Thoonsen, recently appeared on the Comptoir IA podcast to discuss AI innovation, customer operations, and the future of business automation. About the Podcast. Comptoir IA is a leading French podcast focused on artificial intelligence and its applications in business. The show features experts from across the AI industry sharing insights on the latest trends, technologies, and real world implementations. Key Discussion Points. The Evolution of Customer Operations. Maxime shared his vision of how AI is transforming customer service from reactive support to proactive, intelligent assistance. He discussed how businesses can leverage AI to not just respond to customer needs, but anticipate them. Building AGO's AI Platform. As AGO's CTO, Maxime provided insights into the technical challenges and innovations behind our AI powered customer operations platform. He explained how we've designed systems that can understand context, learn from interactions, and continuously improve performance. Our platform has particularly focused on complex context engineering, enabling AI agents to maintain deep understanding of multi turn conversations and intricate customer scenarios. The Future of AI in Business. The conversation covered exciting developments in AI technology and how they're reshaping business operations. Maxime discussed the importance of building AI systems that complement human expertise rather than replace it. Practical AI Implementation. Drawing from AGO's experience with enterprise customers, Maxime shared practical advice for organizations looking to implement AI solutions effectively and responsibly. ### MCP UI: A new way to build interactive AI-powered experiences Date: 2025-09-10. Shopify lit the fuse on MCP UI with their "MCP UI: Breaking the text wall with interactive components" blog post, published on August 5, 2025. That post lays out how MCP UI extends the Model Context Protocol to return embedded interactive components. As Shopify explains: "Commerce UI is deceptively complex, we thought, what if MCP could return not just data, but fully interactive UI components?" The problem before MCP UI. Traditional LLM driven UIs rely on extended markdown (with custom tags or placeholders), search and replace patterns (embedding videos, links, or components), and component injection (turning placeholders into React/HTML components). But these methods have two drawbacks. LLMs struggle to generate complex UIs reliably, often producing incomplete or malformed instructions. Unlike function calls, which have a clear structure, UI generation is more free form and error prone. Rendering UIs on the frontend is tightly coupled to the specific backend and frontend implementations preventing reuse across third parties applications. How MCP UI renders complex third party UIs reliably. Instead of incomplete instructions, MCP UI returns a fully generated HTML component that the frontend can display directly. The process: 1. The LLM selects a tool to call with appropriate data, for example DisplayProduct({ product_id: 22 }). 2. The backend gets the template linked to the tool. 3. Then it renders the HTML using a templating engine like Jinja. 4. The output is sent to the frontend as a complete HTML string of the component. 5. On the frontend, this HTML is rendered inside a sandboxed iframe for security and stability using a MCP UI renderer. With this architecture, third parties can host their own MCP UI servers, enabling a decentralized ecosystem of interactive components. For example, a payment processor could host a checkout component that any LLM powered app could embed. Philosophically, this is similar to how MCP enables decentralized data retrieval. It also enables richer interactions like sorting and filtering, things Markdown can't handle. This complements LLMs well, since they're notoriously poor at sorting long lists by generated criteria. Why This Matters for Commerce. AI assistants like ChatGPT and Perplexity are moving into e commerce. They're no longer just answering questions, they're becoming storefronts. Text only chat is insufficient for retail. Customers expect product images, size and color selectors, dynamic pricing and availability, interactive shopping carts. MCP UI delivers exactly that: immersive shopping experiences directly inside AI assistants. If your backend doesn't support something like MCP UI you're about to lose share to competitors who let customers shop directly through AI. Customer Service With Actions. Beyond shopping, MCP UI also transforms customer service. Instead of static responses, assistants can present actionable UI components: buttons to reschedule a delivery, forms to request a return or refund, interactive troubleshooting flows with dynamic updates, secure payment or authentication prompts. This shifts customer service from being reactive and text heavy into a guided, interactive experience where problems get solved inside the conversation itself. The assistant is no longer just a support agent, it's an operational front end able to trigger real backend actions seamlessly. This is exactly what we do at AGO with our MCP UI integration. Our AI agents create dynamic, interactive UIs that let customers solve their issues directly within the chat. ### Why Chatbots Fail on Complex Tickets (And What to Do About It) Date: 2026-02-15. There is a pattern we see every time a company deploys a chatbot for customer support. The first demo goes great. Simple questions get good answers. The team is excited. Then real tickets start flowing in. The 40% wall. A customer writes: "I was charged twice for my last order, and I also want to change the shipping address on my next order." Two problems in one message. The chatbot picks one, ignores the other, and the customer writes back frustrated. We tracked this across several AGO deployments. On average, 35% of support tickets contain more than one intent. And when a traditional chatbot hits one of these multi intent messages, the resolution rate drops from around 80% to below 40%. That is a massive gap. And it explains why many companies end up with their chatbot handling only the easy questions while humans still do the heavy lifting. What actually breaks. It is not that the AI cannot understand the second question. The problem is architectural. Most chatbot frameworks are built around a single request/response loop. The user sends a message, the bot picks an intent, calls one tool, and responds. When the user packs two requests into one message, the bot has to pick. And it almost always picks the first one, because that is what appears first in the text. Even if you build intent splitting into the pipeline, you run into a second problem: the two tasks might depend on each other. The customer wants a refund on the duplicate charge and wants to know if the refund changes the total on their next order. You cannot answer the second question without completing the first. How persistent agents handle this differently. A DeepAgent does not work in a single request/response loop. It receives the full message, breaks it into sub tasks, and executes each one sequentially while keeping the full context. For the example above, the agent would: 1. Look up the order and confirm the duplicate charge. 2. Process the refund. 3. Check the next order and recalculate the total with the refund applied. 4. Update the shipping address on the next order. 5. Reply to the customer with everything resolved in one message. The key difference: the agent maintains state across all these steps. The refund amount from step 2 is available in step 3. The shipping address change in step 4 can reference the updated order from step 3. What we learned in production. After switching from a traditional chatbot to DeepAgents on multi intent tickets, we saw the resolution rate on those cases go from 38% to 67%. Not perfect, but a meaningful jump. The cases that still fail are mostly ones where the customer references something ambiguous ("the thing I ordered last time") and the agent picks the wrong interpretation. We are working on better disambiguation, but that is a topic for another post. The takeaway: if your chatbot metrics look great on simple questions but your overall resolution rate is stuck, check how it handles messages with more than one request. That is probably where the gap is.