Guillaume "G/" Moigneu
@guillaume.id
Field CTO @ Upsun. I help leaders and teams augment themselves with AI. The practice, not the commentary. I build production systems and explain them to a CEO. Building gaal (open-source agent access layer) and number6.ai. Writing at guillaume.id.
Agent routers coordinate model choice, tool execution, and verification across a workflow.
Price is one routing input alongside completion rate, latency, available context, and verification cost.
That means the router needs to understand the work it is scheduling.
#ai #agents #router
Price is one routing input alongside completion rate, latency, available context, and verification cost.
That means the router needs to understand the work it is scheduling.
#ai #agents #router
July 21, 2026 at 2:37 PM
An LLM can challenge your reasoning or rationalize the decision you already wanted.
The difference is process.
I built Pressure-test decisions to surface assumptions, test uncertainty, and end with a concrete action.
I still make the call. The skill forces me to show my work.
The difference is process.
I built Pressure-test decisions to surface assumptions, test uncertainty, and end with a concrete action.
I still make the call. The skill forces me to show my work.
July 17, 2026 at 9:25 AM
An LLM can challenge your reasoning or rationalize the decision you already wanted.
The difference is process.
I built Pressure-test decisions to surface assumptions, test uncertainty, and end with a concrete action.
I still make the call. The skill forces me to show my work.
The difference is process.
I built Pressure-test decisions to surface assumptions, test uncertainty, and end with a concrete action.
I still make the call. The skill forces me to show my work.
My YouTube research agent has seven workflow steps. Only two stages use an LLM.
Code handles fetching, filtering, ranking, caching, and persistence. Models extract transcript evidence and synthesize patterns.
Put model calls where ambiguity creates value.
Code handles fetching, filtering, ranking, caching, and persistence. Models extract transcript evidence and synthesize patterns.
Put model calls where ambiguity creates value.
July 13, 2026 at 7:37 PM
My YouTube research agent has seven workflow steps. Only two stages use an LLM.
Code handles fetching, filtering, ranking, caching, and persistence. Models extract transcript evidence and synthesize patterns.
Put model calls where ambiguity creates value.
Code handles fetching, filtering, ranking, caching, and persistence. Models extract transcript evidence and synthesize patterns.
Put model calls where ambiguity creates value.
Really good write-up by Fabien Potencier. Challenges ahead for organizations to fully adopt AI development. Best practices, governance, cost-control, etc. We are all trying to adapt without a clear path, which this article tries to map out. https://upsun.com/blog/8-stages-ai-engineering-maturity/
June 12, 2026 at 10:09 AM
Really good write-up by Fabien Potencier. Challenges ahead for organizations to fully adopt AI development. Best practices, governance, cost-control, etc. We are all trying to adapt without a clear path, which this article tries to map out. https://upsun.com/blog/8-stages-ai-engineering-maturity/
6 months running my own AI personal agent and this weekend I hit a wall.
Nothing beats strongly typed structured data.
Markdown-only agents are painful. Slow, dozens of tool calls, tokens burning for nothing. Do I really need to wait 70 seconds to flip a task from In Progress to Done?
Nothing beats strongly typed structured data.
Markdown-only agents are painful. Slow, dozens of tool calls, tokens burning for nothing. Do I really need to wait 70 seconds to flip a task from In Progress to Done?
June 9, 2026 at 9:07 AM
6 months running my own AI personal agent and this weekend I hit a wall.
Nothing beats strongly typed structured data.
Markdown-only agents are painful. Slow, dozens of tool calls, tokens burning for nothing. Do I really need to wait 70 seconds to flip a task from In Progress to Done?
Nothing beats strongly typed structured data.
Markdown-only agents are painful. Slow, dozens of tool calls, tokens burning for nothing. Do I really need to wait 70 seconds to flip a task from In Progress to Done?
A great idea inspired by conventional commits: Conventional Comments
https://conventionalcomments.org/
It's so true that sometimes I don't even know what to do with a comment on a Google Doc. Acknowledge, reply or just close? This would help!
https://conventionalcomments.org/
It's so true that sometimes I don't even know what to do with a comment on a Google Doc. Acknowledge, reply or just close? This would help!
May 25, 2026 at 4:01 PM
A great idea inspired by conventional commits: Conventional Comments
https://conventionalcomments.org/
It's so true that sometimes I don't even know what to do with a comment on a Google Doc. Acknowledge, reply or just close? This would help!
https://conventionalcomments.org/
It's so true that sometimes I don't even know what to do with a comment on a Google Doc. Acknowledge, reply or just close? This would help!
What are you working on this long weeekend? I've just revamped my AI assistant. After trying so many solutions (Hermes, graphs, etc.)
I am back to a way simpler method. Pi + local gemma 4 + markdown files + qmd + Obsidian.
Tasks management, journal and my knowledge aggregations all fit in there.
I am back to a way simpler method. Pi + local gemma 4 + markdown files + qmd + Obsidian.
Tasks management, journal and my knowledge aggregations all fit in there.
May 24, 2026 at 5:35 PM
What are you working on this long weeekend? I've just revamped my AI assistant. After trying so many solutions (Hermes, graphs, etc.)
I am back to a way simpler method. Pi + local gemma 4 + markdown files + qmd + Obsidian.
Tasks management, journal and my knowledge aggregations all fit in there.
I am back to a way simpler method. Pi + local gemma 4 + markdown files + qmd + Obsidian.
Tasks management, journal and my knowledge aggregations all fit in there.
I wanted to build a lead-catching LLM based chatbot but I had to figure out which defenses were actually worth the complexity.
A Chevrolet chatbot agreed to sell a $76K Tahoe for $1 after one prompt injection.
guillaume.id/blog/defendi...
A Chevrolet chatbot agreed to sell a $76K Tahoe for $1 after one prompt injection.
guillaume.id/blog/defendi...
Defending LLM chatbots against prompt injection and topic drift
Practical defenses for LLM chatbots against prompt injection and topic drift, including system-prompt hardening, input validation, output filtering, topic control, and pragmatic architecture choices.
guillaume.id
February 27, 2026 at 3:48 PM
I wanted to build a lead-catching LLM based chatbot but I had to figure out which defenses were actually worth the complexity.
A Chevrolet chatbot agreed to sell a $76K Tahoe for $1 after one prompt injection.
guillaume.id/blog/defendi...
A Chevrolet chatbot agreed to sell a $76K Tahoe for $1 after one prompt injection.
guillaume.id/blog/defendi...
Quickbit: If you're struggling defining priorities and goals (as a person, a company), take a look at Daniel Miessler's Telos framework. Spent the last hour talking to my laptop to build it! It really helps. github.com/danielmiessl...
GitHub - danielmiessler/Telos: Telos is an open-sourced framework for creating Deep Context about things that matter to humans.
Telos is an open-sourced framework for creating Deep Context about things that matter to humans. - danielmiessler/Telos
github.com
February 23, 2026 at 11:50 PM
Quickbit: If you're struggling defining priorities and goals (as a person, a company), take a look at Daniel Miessler's Telos framework. Spent the last hour talking to my laptop to build it! It really helps. github.com/danielmiessl...
Your agents.md is probably hurting more than it helps. A new ETH Zurich study tested coding agents across hundreds of real GitHub issues and found that LLM-generated context files reduced success rates by 3% while adding 20% to inference costs.
AGENTS.md
AGENTS.md is a simple, open format for guiding coding agents. Think of it as a README for agents.
AGENTS.md
February 23, 2026 at 4:11 PM
Your agents.md is probably hurting more than it helps. A new ETH Zurich study tested coding agents across hundreds of real GitHub issues and found that LLM-generated context files reduced success rates by 3% while adding 20% to inference costs.
I love initiatives like Claude Code Security (even if it's gated for now). But I'm also terrified of them.
claude.com/solutions/cl...
claude.com/solutions/cl...
Claude Code Security | Anthropic by Claude
From scan to fix, seamlessly. Claude scans your codebase for vulnerabilities, validates findings, and recommends patches you can review and approve.
claude.com
February 21, 2026 at 5:33 PM
I love initiatives like Claude Code Security (even if it's gated for now). But I'm also terrified of them.
claude.com/solutions/cl...
claude.com/solutions/cl...
Average typing speed: 40 WPM. Conversational speech: 130-150 WPM. Your fingers literally can't keep up with your brain. Voice typing fixes that gap. SuperWhisper nails this on macOS. Press a hotkey, talk, text appears. Accurate, offline, fast. But it's Mac-only with no Linux port planned.
February 20, 2026 at 5:31 PM
Average typing speed: 40 WPM. Conversational speech: 130-150 WPM. Your fingers literally can't keep up with your brain. Voice typing fixes that gap. SuperWhisper nails this on macOS. Press a hotkey, talk, text appears. Accurate, offline, fast. But it's Mac-only with no Linux port planned.
I am not a big fan of OpenClaw (yet) but I find tremendous value in automating simple tasks.
I've worked yesterday and my own terminal based assistant: n6. In honor of the beloved Caprica Six.
Extensible python terminal utility wrapping Claude.
github.com/gmoigneu/n6
I've worked yesterday and my own terminal based assistant: n6. In honor of the beloved Caprica Six.
Extensible python terminal utility wrapping Claude.
github.com/gmoigneu/n6
February 19, 2026 at 4:11 PM
I am not a big fan of OpenClaw (yet) but I find tremendous value in automating simple tasks.
I've worked yesterday and my own terminal based assistant: n6. In honor of the beloved Caprica Six.
Extensible python terminal utility wrapping Claude.
github.com/gmoigneu/n6
I've worked yesterday and my own terminal based assistant: n6. In honor of the beloved Caprica Six.
Extensible python terminal utility wrapping Claude.
github.com/gmoigneu/n6
I ran react-doctor on a real production codebase. 88 out of 100. 4 errors across 139 files. Better than I expected and the 4 errors are exactly the kind that bite you in production.
February 18, 2026 at 11:23 PM
I ran react-doctor on a real production codebase. 88 out of 100. 4 errors across 139 files. Better than I expected and the 4 errors are exactly the kind that bite you in production.
I'm starting a new AI Weekly series. I was doing that for my own needs but I think it's worth sharing. If you have any good source to add to my tracker let me know!
www.linkedin.com/pulse/weekly...
www.linkedin.com/pulse/weekly...
Weekly AI Review - 2026-02-17
Anthropic closed a staggering $30 billion Series G at a $380 billion valuation. The second-largest venture deal in history as Claude Code's revenue doubled to $2.
www.linkedin.com
February 18, 2026 at 2:10 PM
I'm starting a new AI Weekly series. I was doing that for my own needs but I think it's worth sharing. If you have any good source to add to my tracker let me know!
www.linkedin.com/pulse/weekly...
www.linkedin.com/pulse/weekly...
My personal take on how to make coding agents reliable.
For me it's all about verification.
The limit isn’t the AI. It’s your infrastructure. Build the verification layer, and the autonomy follows.
The full article: devcenter.upsun.com/posts/making...
For me it's all about verification.
The limit isn’t the AI. It’s your infrastructure. Build the verification layer, and the autonomy follows.
The full article: devcenter.upsun.com/posts/making...
Making coding agents (Claude Code, Codex, etc.) reliable
The bottleneck for AI coding agents isn't model capability. It's your verification infrastructure. Here's how to fix that.
devcenter.upsun.com
February 18, 2026 at 2:08 PM
My personal take on how to make coding agents reliable.
For me it's all about verification.
The limit isn’t the AI. It’s your infrastructure. Build the verification layer, and the autonomy follows.
The full article: devcenter.upsun.com/posts/making...
For me it's all about verification.
The limit isn’t the AI. It’s your infrastructure. Build the verification layer, and the autonomy follows.
The full article: devcenter.upsun.com/posts/making...
You know that moment when Claude forgets you use pnpm? Again. For the third time today. I got tired of repeating myself. And then I found napkin.
February 11, 2026 at 4:00 PM
You know that moment when Claude forgets you use pnpm? Again. For the third time today. I got tired of repeating myself. And then I found napkin.
If you want the slides from my hashtag#AIDay talk about making coding agents reliable, here you go: guillaume.id/pdfs/2026-02...
And the 8 pillars verification checklist: gist.github.com/gmoigneu/a96...
Thank you Anne-Sophie Norca from dotAI by dotConferences for organizing this!
And the 8 pillars verification checklist: gist.github.com/gmoigneu/a96...
Thank you Anne-Sophie Norca from dotAI by dotConferences for organizing this!
February 10, 2026 at 2:46 PM
If you want the slides from my hashtag#AIDay talk about making coding agents reliable, here you go: guillaume.id/pdfs/2026-02...
And the 8 pillars verification checklist: gist.github.com/gmoigneu/a96...
Thank you Anne-Sophie Norca from dotAI by dotConferences for organizing this!
And the 8 pillars verification checklist: gist.github.com/gmoigneu/a96...
Thank you Anne-Sophie Norca from dotAI by dotConferences for organizing this!
What primitives will replace human authorship and review as the foundation of trust?
Teams will track "code reading coverage" metrics. But here's the kicker: they'll strategically drive it DOWN in safe areas.
Teams will track "code reading coverage" metrics. But here's the kicker: they'll strategically drive it DOWN in safe areas.
February 10, 2026 at 8:26 AM
What primitives will replace human authorship and review as the foundation of trust?
Teams will track "code reading coverage" metrics. But here's the kicker: they'll strategically drive it DOWN in safe areas.
Teams will track "code reading coverage" metrics. But here's the kicker: they'll strategically drive it DOWN in safe areas.
Reposted by Guillaume "G/" Moigneu
We’re heading to #AIDay2026! 🤖 Guillaume Moigneu will be on stage talking coding agent readiness. Can't attend? Don't worry, the tech track will be livestreamed 👀 Sign-up here: luma.com/dotAI-techtr...
📍 Paris, France | ⏰ 12:00–12:15 pm CET 👉 ai-day.org/programme/
#AIDay2026 #Upsun #AIEngineering
📍 Paris, France | ⏰ 12:00–12:15 pm CET 👉 ai-day.org/programme/
#AIDay2026 #Upsun #AIEngineering
February 5, 2026 at 4:00 PM
We’re heading to #AIDay2026! 🤖 Guillaume Moigneu will be on stage talking coding agent readiness. Can't attend? Don't worry, the tech track will be livestreamed 👀 Sign-up here: luma.com/dotAI-techtr...
📍 Paris, France | ⏰ 12:00–12:15 pm CET 👉 ai-day.org/programme/
#AIDay2026 #Upsun #AIEngineering
📍 Paris, France | ⏰ 12:00–12:15 pm CET 👉 ai-day.org/programme/
#AIDay2026 #Upsun #AIEngineering
Discussing with industry old timers friends yesterday we came up to the same conclusion. Everything is so exciting right now like it was in 1999.
At the time I was discovering PHP and started to publish a lot of different projects with my trusty Dreamweaver. But I was feeling so empowered.
At the time I was discovering PHP and started to publish a lot of different projects with my trusty Dreamweaver. But I was feeling so empowered.
February 6, 2026 at 9:28 AM
Discussing with industry old timers friends yesterday we came up to the same conclusion. Everything is so exciting right now like it was in 1999.
At the time I was discovering PHP and started to publish a lot of different projects with my trusty Dreamweaver. But I was feeling so empowered.
At the time I was discovering PHP and started to publish a lot of different projects with my trusty Dreamweaver. But I was feeling so empowered.
I've always been an advocate of data privacy. And even if we have to put ourselves for professional reasons, we always need to be mindful of the potential risks of doing it.
Check out Carole Cadwalladr talk at TED. Worth every second of your time!
www.ted.com/talks/carole...
Check out Carole Cadwalladr talk at TED. Worth every second of your time!
www.ted.com/talks/carole...
This is what a digital coup looks like
“We are watching the collapse of the international order in real time, and this is just the start,” says investigative journalist Carole Cadwalladr. In a searing talk, she decries the rise of the “bro...
www.ted.com
April 14, 2025 at 3:32 PM
I've always been an advocate of data privacy. And even if we have to put ourselves for professional reasons, we always need to be mindful of the potential risks of doing it.
Check out Carole Cadwalladr talk at TED. Worth every second of your time!
www.ted.com/talks/carole...
Check out Carole Cadwalladr talk at TED. Worth every second of your time!
www.ted.com/talks/carole...
Hello everyone! I wanted to share some thoughts on JD Vance's recent speech at the France AI Summit. I'm really concerned about his stance on regulation.
#AISafety #AIRegulation #FranceAISummit #AIEthics #ProfitWithPurpose #AI
#AISafety #AIRegulation #FranceAISummit #AIEthics #ProfitWithPurpose #AI
February 17, 2025 at 3:05 PM
Hello everyone! I wanted to share some thoughts on JD Vance's recent speech at the France AI Summit. I'm really concerned about his stance on regulation.
#AISafety #AIRegulation #FranceAISummit #AIEthics #ProfitWithPurpose #AI
#AISafety #AIRegulation #FranceAISummit #AIEthics #ProfitWithPurpose #AI
🔥 Vous êtes invité(e) à notre petit-déjeuner pour découvrir comment la durabilité crée un avantage concurrentiel et attire les investissements dans l'écosystème de l'IA⚡️
brnw.ch/21wQitQ
📅 Date: 11 February, 2025
🕗 Time: 09:00 - 11:00 AM
📍 Location: Digital Village Paris
brnw.ch/21wQitQ
📅 Date: 11 February, 2025
🕗 Time: 09:00 - 11:00 AM
📍 Location: Digital Village Paris
February 2, 2025 at 4:47 PM
🔥 Vous êtes invité(e) à notre petit-déjeuner pour découvrir comment la durabilité crée un avantage concurrentiel et attire les investissements dans l'écosystème de l'IA⚡️
brnw.ch/21wQitQ
📅 Date: 11 February, 2025
🕗 Time: 09:00 - 11:00 AM
📍 Location: Digital Village Paris
brnw.ch/21wQitQ
📅 Date: 11 February, 2025
🕗 Time: 09:00 - 11:00 AM
📍 Location: Digital Village Paris