iiinova — where your digital self keeps evolving

For people using Getting Things Done out there. Remember context. I mentioned high enzgegy / low energy as a context. I think now I can add high tokens and low/no tokens to the tasks. Tokens budgets is a thing 🙃

That’s the intermediary state yes. I was assuming they could already by hand but expensive. It will just be cheaper for state actors. And affordable for big orgs. But soon the discovery bar will be high again when vendors check their software against Mythos before release.

Les Cast Codeurs @lescastcodeurs.com · 29/04/2026

Reposted by Emmanuel Bernard

Episode 339 - Clotûre de Devoxx France 2026 par les cast codeurs @DevoxxFr #PolosRouges sur youtube.com/watch?v=nT-kisRPfoA et en podcast lescastcodeurs.com/2026/04/29/lcc-3…

093

Also reading between the lines, cybersecurity is just the domain that anthropic decided to apply mythos on. Other domains will follow and thinking about price per value will be more important than ever

201

- as a non security person, you could think security researcher are out of a job but I think they can do more and will be asked to do more.

100

- mythos project evaluation runs are reportedly in the order of 10k for analysys and build of the main exploit, so you can’t run in each PR shift left style :) - AI is becoming more expensive as it gets more useful.

100

- I think this means that software will be more expensive, not less and that an software editor brings value (sharing the cost around « coding ». I base that on the security aspect and others I am seeing

100

- Anthropic happened to have reached this level and they had to announce it now to reap the benefit of « thousands of vulnerabilities found », the next contender will get what remains (low)

- mythos will be expensive. I think 10x Opus 4.6. So above script kiddy but affordable for a big enough org - it is conveniently good marketing for Anthropic IPO (see later point)

My take on the Claude Mythos claims (reportedly very good at offensive cybersecurity) - I think the claims are legit - the industry catch up to fix and raise to the level will be somewhat painful

Emmanuel Bernard @emmanuelbernard.com · 25/04/2026

So I guess you could use goal to decide the next task in my model but I don’t think that’s appropriate as it’s a simple hierarchical DAG with multiple roots.

Emmanuel Bernard @emmanuelbernard.com · 25/04/2026

What I described was more task addition to an existing graph describing blocking and dependent tasks. New tasks created *by the subagents themselves*. Then some solver decides what is the next task.

210

Emmanuel Bernard @emmanuelbernard.com · 25/04/2026

So GOAL kinds of solvers have a LEGO set of operations with pre post condition and input / output “types”. So yes based on where they are and the state, the solver choses the next Lego block.

Guillaume Laforge @glaforge.dev · 23/04/2026

Reposted by Emmanuel Bernard

Merci à tous ceux qui sont venus à @devoxx.fr nous voir @emmanuelbernard.com et moi pour parler des protocoles pour les agents IA, avec le focus sur #MCP et #A2A, et implémentations en #ADK et @langchain4j.dev Comme promis, les slides de notre présentation : speakerdeck.com/glaforge/sta...

speakerdeck.com

Standards et agents IA : un tour d’horizon de MCP, A2A, ADK et plus encore

Les fondations des agents IA reposent sur une poignée de protocoles communs que vous devez maîtriser pour tirer le meilleur parti de votre LLM et de vot…

1224

Emmanuel Bernard @emmanuelbernard.com · 24/04/2026

Non le GOAP est statique de ce que je comprends dans sa définition de graphe. Là il y a une émergence de dependence de tâches qui peuvent être ajoutées par les agents eux même “on the fly”. Enfin si c’est la conversation dont the me souvient 🤣

100

Reposted by Emmanuel Bernard

Damien CACHEUX @damcx.bsky.social · 22/04/2026

Standards et protocoles des agents IA avec MCP, A2A avec @emmanuelbernard.com et @glaforge.dev @devoxx.fr #DevoxxFR

051

Reposted by Emmanuel Bernard

Rémi Verchère ❄️ @r.verchere.fr · 22/04/2026

J’ai attendu sagement l’interview de @emmanuelbernard.com, qui discutait avec @lours.me pour #devoxxfr studio

182

Emmanuel Bernard @emmanuelbernard.com · 20/04/2026

Pour le studio @DevoxxFR, j’ai interviewé @lau_devil sur notre relation à l’IA, les emotions et le rôle crucial de l’éducation youtu.be/-RURogH7PCY?... Très belle rencontre qui m’a bien challengé. La suite a @devoxx.fr cette semaine !

youtu.be

L'IA : ange 😇 ou démon 😈 ?

YouTube video by Devoxx France

185

Emmanuel Bernard @emmanuelbernard.com · 14/04/2026

I like to chastise Sonnet and Opus for their mistakes from time to time. "Ground your answer, this is wrong". I want to show them, they haven't mastered it all... yet.

020

Emmanuel Bernard @emmanuelbernard.com · 09/04/2026

Sharing mindmaps with markdown emmanuelbernard.com/blog/2026/04...

emmanuelbernard.com

Emmanuel Bernard @emmanuelbernard.com · 09/04/2026

Remember the "Fire everyone, keep the coders!" cartoon. That did not age too well.

020

Emmanuel Bernard @emmanuelbernard.com · 02/04/2026

it works from here. Maybe a temp glitch.

121

Emmanuel Bernard @emmanuelbernard.com · 29/03/2026

Oui ! On a la plateforme normalement et on doit configurer tout ça. Bref c’est en cours.

040

Read the thread above @swallez.com it was my response to your message.

Crap I failed to reference @swallez.com and his post bsky.app/profile/swal...

210

So what you propose should not be called a spec. It’s closer to code. And I don’t see English as being a good basis for it.

100

Whether it’s code or your advanced spec Sylvain (of which I have doubts about), humains won’t read it much in the future. Like I don’t read much assembly code today.

It’s quite uncanny and what defines these tools vs the other technologies before it. It makes some (stochastic and algorithmic based) judgment calls.

100

What is fascinating is that coding agents can adjust one based on the other and iterate on them like a pyramid. And coding agents do “fill the blank”, the unknown, the unsaid.

100

AGENTS.md is a simple, open format for guiding coding agents. Think of it as a README for agents.

But there is something to the cascading for artéfacts. A WHY artefact not to lose the big picture (say a readme), the HOW (agents.md, architecture.md) as a constraining set of requests (I say requests because the coding agent can ignore them), the high level WHAT (spec.md) and the real WHAT (code).

agents.md

AGENTS.md

100

So to describe the what something like code ends up being better. And between English code and Java code, I know what I prefer to read, understand and navigate. Java is less expressive and more deterministic and syntactically structured. These constraints make it a good descriptor of WHAT.

100

Lean and you leave loads of undefined semantic. Detail to fix that and English (let alone French) will turn your text into a very long, verbose and hard to parse text. That’s because human language has lots of uncertainty built in and fighting it falls under its own weight.

100

Emmanuel Bernard @emmanuelbernard.com · 15/03/2026

I don’t agree with you Sylvain and I think the article does not either. People that have written specs that need to be implemented by several other systems, that spend countless hours to refine it, know that uncertainty and unclarity creeps in at every opportunity.

111

Comme dans beaucoup d’endroits, ma ville n’a que des listes citoyennes non affiliés. Mais les maires et certains conseillés sont grands électeurs pour les sénatoriales. C’est un trou dans la raquette de la représentation nationale.

012

Reposted by Emmanuel Bernard

Devoxx France @devoxx.fr · 09/03/2026

Venez vivre un Super Mercredi à Devoxx France 2026 🙌 Une offre spéciale pour la journée du mercredi 22 avril : 250 € au lieu de 370 € avec le code "SuperMercredi26" ⚠️ 300 places seulement 👉 reg.devoxx.fr

146

Emmanuel Bernard @emmanuelbernard.com · 08/03/2026

Now. At least I saw it only today.

Emmanuel Bernard @emmanuelbernard.com · 08/03/2026

oh looks like Claude code can not react to messages you send it while it is thinking. That's going to be big.

100

Emmanuel Bernard @emmanuelbernard.com · 06/03/2026

What do people use to clean up macOS unexpected disk space usage ? Including slack, Chrome, Firefox etc caching Nowadays any random app believes it can use 10GB for caching to be fast. Well you’re not alone on my machine buddy.

431

Emmanuel Bernard @emmanuelbernard.com · 06/03/2026

Boy! Watch for Claude Desktop and Docker sandbox eating your hard drive alive. I did not understand why my incremental backup was eating 4TB of space in a matter of a couple of weeks. Now I know.