Vincent Dal Maso

Founder/CTO, Edko + Buildtrackr.
Running agentic coding through real production fire on two SaaS platforms.
AI as leverage, not a replacement.

[Vincent Dal Maso @vincent_dalmaso]

Alignment Under Pressure

That’s the simplest way to describe how I work

I’m interested in systems that hold under real constraints:
- messy inputs
- fragmented workflows
- unclear ownership
- operational noise

My work is turning a messy reality into a deployable systems

[Vincent Dal Maso @vincent_dalmaso]

Verification is half the setup.
The other half is where the rules live.

A check the agent can rewrite is a suggestion.
A check outside its reach is a gate.

A simple and boring first step in that direction:

add yourself to `CODEOWNERS` on eslint/tsconfig/pyproject, CI workflows files and checks/. + Branch protection on.

The agent can read every rule. Changing one needs my approval.

Most people build the checks/verifications and skip the locks.

[Vincent Dal Maso @vincent_dalmaso]

Agreed, and it changes what a code review is for.

If observing and refactoring got cheap, the scarce thing is no longer reading the code line by line, it's knowing whether the result is right.

And a nice Grok @bot running a routine to check twice a day if any issue in production surface in the logs/monitorings and spin up agents to fix them is a real cheat code to reduce the pressure.

[dax @thdxr]

the slop conversation is so one sided, it's not just that you can produce more code

you can also observe, debug, and refactor more code

even if the code is slop you have such good tools to deal with it. even if you write code manually, it's still 100x easier to maintain it

[Vincent Dal Maso @vincent_dalmaso]

One layer on top of "codebase is memory": memory only helps if the agent can't quietly erase it.

If the same agent that writes the code can also loosen the lint config, skip a test, or tweak the CI step, the constraint erodes the first time it gets in the way.

So I keep the gates outside its write path: it can read them, not change them. Every miss becomes a new gate. The tradeoff is friction: some rules will be wrong and block good work, and fixing them should stay a human call.

That small cost is what keeps the memory honest.

[lauren @poteto]

something that keeps bothering me is how some people see slop as something that is happening to them, like they're some kind of helpless bystander powerless to do anything about it

excuse me but you are a software engineer, you literally learned how to make sand think and make computers do your bidding?! and now you can't do anything about it? please

the 2 things you should know to cut through all the noise:

1. you need to give your agent a way to verify its own work (i just posted about this so won't rehash)
2. your codebase is memory

the 2nd is not new either, we have always had lint rules, compilers with good diagnostics and strong type systems, LSPs, etc to help us write better code. there is no difference when agents write the code instead, you just need more of them, and stronger ones.

forget rules and skills. you can add those later.

yes, your codebase is a form of memory to agents! agents love extending what already exists. so every nit and every annoyance you've ever seen, every mistake and bad pattern you observe from agents, needs to turn into a constraint, or the agents will just keep repeating those same mistakes like a virus in your code. then you keep observing and repeating this process until your codebase is locked down enough that even agents with little context can one shot a high quality PR.

an even better direction is to refactor/rewrite your code such that agent anti-patterns become categorically impossible. use type systems to your advantage, use libraries/frameworks that reduce the architectural search space, or heck, create your own framework like we did for grok bot and design high quality primitives and well thought out APIs

[Vincent Dal Maso @vincent_dalmaso]

The practical version: treat every bad test as a missing rule, then ask which layer should own it.

Style or structure issues go to lint.
Behavior gaps go to a test that fails on the bad version first.
Judgment calls, like testing the mock instead of the code, go to a review skill that runs on every change.

A line in the system prompt is advice the agent can forget under pressure.

A rule in the codebase is enforced on every run, for every agent, including the next model you switch to.

[Matt Pocock @mattpocockuk]

Don't blame the agent for writing the bad test

Blame the codebase for allowing the bad test to be written

[Vincent Dal Maso @vincent_dalmaso]

https://t.co/RV3fxd8NTa

[Treg @treg_ai]

Introducing OpenSource Signal Monitor

- Monitor live web for any buyer signals
- Route to 80+ vendors, best win
- 85% cheaper than clay

Try it at https://t.co/PK8URwe2I3

Repo link below 👇

[Vincent Dal Maso @vincent_dalmaso]

La version gratuite de ChatGPT ne donne accès que à 5 % de ce que l'on peut faire avec les agents.

Le monde de la conversation comme GPT3 en 2023 est révolu depuis novembre 2025.

Les agents ont pris le pas et ce qu'ils peuvent faire est infini. La technologie (code/hardware) était limitée ces cinq dernières années par notre capacité à inventer/créer et le temps disponible. Aujourd’hui, la contrainte du temps est terminée.

Je rejoins @StephaneMallard sur l'avenir que l'IA nous réserve. Je pense que la seule chose qu'il nous restera sera notre conscience de soi, notre capacité à penser et à créer (jusqu’à ce que l'on soit complètement dépassé).

[Vincent Dal Maso @vincent_dalmaso]

@StephaneMallard Je suis dessus depuis cet été.
La suite logicielle + le modèle c'est vraiment dingue

[Vincent Dal Maso @vincent_dalmaso]

@techandsundry Grok bot config is the new config hell for developpers.

look at my grok bot config is the new loon at my neovim config

[Vincent Dal Maso @vincent_dalmaso]

@GergelyOrosz And Gate/Rules verification too.

[Vincent Dal Maso @vincent_dalmaso]

@leerob @bot Be able to assign/disable plugins(accounts) by bots.
so one bot has one job and only reach to the accounts/plugins allowed !

[Vincent Dal Maso @vincent_dalmaso]

@brian_armstrong We'll see less UI.

More API. Most of the work will be programs agents write, or agents calling MCP and the API directly

[Vincent Dal Maso @vincent_dalmaso]

@poteto I've started using IRMPEW to catch gaps between me and my agent. No data yet. Promising though
https://t.co/bAnKSeBhGA

[Vincent Dal Maso @vincent_dalmaso]

This summer I told a friend that I am incredibly enthusiastic about AI and the wonders it will give us.

He told me “You want to know what I think ?” deep inside I was not expecting any of what came out of his mouth.

“We, humans, are a virus and we are going to kill this earth”.

First thing that crossed my mind was “what the f*ck !?!”. I knew he was into eco, low tech and stuff. But what in hell could put such thoughts in his mind ?

Second, he has a son, 5 years old like mine and I said to myself: “If he thinks his own life is a disease to this world, what stories, could he be telling his son ?”

As parents we have the chance to give our children the spirit, the hope, the strength and arms to deal with the world we know and the world they will encounter on their own.

The people spreading those messages are actually burning the soul of humanity. They think they do the right thing but it is turning into the biggest crime we ever seen. It counts in millions. Babies not born from parents without hope. Millions of suicides by people that couldn’t see the light in the future. Millions of minds plagued by this messages.

The future is bright, not without obstacles but they build the humans we are and we will be. We must keep up the torch for those who have lost sight of the way humans have always walk this earth. One step and problem at a time.

[DHH @dhh]

The constant decel messaging in Europe is insufferable as an adult, but its effect on kids is worse. As a parent, it's your obligation to counterprogram this nonsense. Show them robots, rockets, and wonders of AI. Show them cathedrals and moon landings. Instill hope in progress.

[Vincent Dal Maso @vincent_dalmaso]

@MatznerJon Greg make money selling ideas he does have to be present when shit hit the fan cause claude does mot fix everything.

[Vincent Dal Maso @vincent_dalmaso]

@marc_ibrahim helldivers 2 🙌

[Vincent Dal Maso @vincent_dalmaso]

I don't use plan mode.

I ask for a plan out loud, then the agent drills the project with me until the shape is right.

[Vincent Dal Maso @vincent_dalmaso]

@AmyNoteApp @mattpocockuk Yes i relate to that. but overall the claude code is bloated in the sense of features and tools.

Opencode 2 seems great, pi is the extreme of that.

I wonder if they will make the choice to shrink claude code with the addition of plugins.

[Vincent Dal Maso @vincent_dalmaso]

Eval score can rise while real work gets worse.

I keep gates the agent cannot rewrite.