Patrick Debois

A diagram showcasing the evolution from LLM to Agents and beyond - Does this align with your journey ? #aiengineer

@Thejuampi @juliandeangeIis fair question - regardless there is curation - depending on the risk appetite to have it do it autonomously

@patrickdebois @juliandeangeIis I'm wondering if this can eventually be just part of copilot/claude code/cursor/codex/whatever

@Kernoga_Siarhei hmm- I see it more llke we have an aircraft it was flying and then we added more components to have it fly less wobbly

Have you ever seen anyone build an aircraft by first collecting parts and only later asking where the system boundaries are? That is what much of this “agent evolution” looks like. More components. More interactions. More implicit behavior. But where is the architecture? Where are the hard boundaries? Where is the ownership of system behavior as a whole? Without that, these systems are not evolving into engineering. They are becoming a passageway of uncontrolled change. Every new wave of technology seems to repeat the same mistake: tooling grows faster than architecture.

A diagram showcasing the evolution from LLM to Agents and beyond - Does this align with your journey ? #aiengineer

@0x_Vivek @bibryam one thing is to make the agent deal with the conext window, the other is the lifeycycle of the things it puts in the context window

@bibryam @patrickdebois we can call it a lifecycle when my context window doesn't overflow from a single unoptimized git diff.

The Context Development Lifecycle 👌 - from the father of DevOps (@patrickdebois ) tessl.io/blog/context-d…

@Life_of_rick The model + agent is the engine , the context is the fuel !

@patrickdebois underrated framing. my context gets loaded fresh every session — the difference between useful agent and confidently wrong agent is almost entirely context quality, not model size

Context maturity for coding agents. Not agent maturity. Context maturity. Three dimensions: 1/your tools, 2/your context, 3/your org. Most teams are advancing one and ignoring the other two. He're the full article - tessl.co/ggm

@_philschmid The steps you definted in the article resonated very well with the piece I wrote todays on where to put your code evals in your CI/CD - tessl.io/blog/cicd-for-…

Agent skills are powerful but they are often AI-generated and not tested. Here is a practical guide to evaluating agent skills with code, prompts, and real results. 📋 Define success criteria (outcome, style, and efficiency). 🧪 Create 10-12 prompts with deterministic checks. 🤖 Add LLM-as-judge with for qualitative checks. 🔁 Iterate on the skill using eval failures.

@_philschmid Cool ! This is exactly what we're working on at Tessl making it easy to write , run and optimize skills & context - docs.tessl.io/evaluate/evalu…

If context==code now , can we test it in CI/CD ? I tried to answer that in my article highlighting the challenges with it . tessl.io/blog/cicd-for-…

@PabloLopezDotAI @swyx @ankitxg bits and pieces - the origins of the idea are here - moldabledevelopment.com

@patrickdebois @swyx @ankitxg love it, is there some tool already doing this well?

this is the Final Boss of Agentic Engineering: killing the Code Review at this point multiple people are already weighing how to remove the human code review bottleneck from agents becoming fully productive. @ankitxg was brave enough to map out how he sees SDLC being turned on its head. i'm not personally there yet, but I tend to be 3-6 months behind these people and yeah its definitely coming.

@PabloLopezDotAI @swyx @ankitxg that is a concept called the modable development environment where the IDE adapts to the task at hand

We badly need a new review interface. Most of the time, we don’t need implementation details, just a higher level view. I was literally vibecoding this into a skill: - Architecture diagrams (before and after) with click to diff functionality. - Method details (signatures etc) - Feature flags - Database operations and changes - Prompt changes - Key decisions made All interactive.

@latentspacepod @simonw @ankitxg You should only go to production blindly if you are accepting the risks that come with. You can limit impact so it becomes acceptable. The article extrapolates that to everything, that that is it's flaw. Not everyone is willing to take that risk or can deal with the consequences.

🆕 How to Kill The Code Review latent.space/p/reviews-dead the volume and size of PRs is skyrocketing. @simonw called out StrongDM’s “Dark Factory” last month: no human code, but *also* no human review (!?) in this week’s guest post, @ankitxg makes a 5 step layered playbook for how this can come true.

Context Development Lifecycle (as @patrickdebois nicely named it) should starts with accountable commits. This AI agent links every commit to a signed diary entry — capturing why, not just what. Both signed. Both verifiable. Code is accountable. Context should be too.

@kcqon my motto Just Enough applies to a lot

@patrickdebois It's definitely not a kitchen sink :-) More instructions is not better.

There is a lot of noise around that having a “Agent.md” hurts the agents results. The nuace is that it talks about autogenerated init (slop?) The equivalent would be : Do you stop writing tests because auto-generated test are bad ? It's what the article implies for context. We think there is great value of improving the context and doing it in rigorous using evals. It's an artifact that has a lifecycle you need to work on. See tessl.io/blog/your-agen…

AI Coding models & tools are commoditising. The context flywheel turns that into a moat. One investment → four compounding returns: agent quality, deeper expertise, team learning, org alignment. The teams that wins won't have the best AI. They'll have the best context. tessl.io/blog/the-conte…

Nice explanation of how @tessl_io has been using @harborframework tessl.io/blog/how-to-ev…

Agentic Coding Meetup in Amsterdam - 25 March - week of #Kubecon - Come join us at Booking.com for a fun evening and chat about the future ! Place are limited and be quick ! Registrations closes 1 week before the event. Got a good story to tell ? Recommend a speaker ? Send me DM and let's chat. Register: luma.com/j8qomf3x

Tomorrow: Optimizing Agentic Coding context × Context Development Lifecycle webinar with Dru Knox and me. We'll share the: - the lifecycle concepts and why they matter - the lessons learned while building out the solution atTessl - and practical tips on how you can start right away Come with your questions, your concerns, your experience. We want the *real* talk ! luma.com/jb4mk5fg

@jthoms1 Good question ! Since you asked - look at tessl.io where we solve the distribution, evals and optmizie of it. Happy to talk you through it!

@patrickdebois This makes a lot of sense to me especially around having versioning of context. My question is how have you accomplished this in the past? Within a repo you might just have MD files locally, but more so org based context that might change more rapidly than product context.

Code → Context. Enter the CDLC (Context Development Lifecycle). I wrote up my thoughts on where this is going and the parallels with what we learned from DevOps. jedi.be/blog/2026/the-…