adlrocha

We were used to say that "if something can be exploited it will be exploited".

Always design tasks delegated to an agent with the feedback loop in mind so they can work autonomously for longer and verify their own work

I love being the dumbest person in the room, and it’s happening a lot lately

At least we did to things right the first time > Josh Payne: Second-time founders be like… > GTM > PMF > Go fully remote > Profitability is king > Must have a Big TAM > Outsource non-core tasks > Have great advisors/investors > Prioritize customer convers…

I am increasingly convinced that moving forward you won’t to find a job, but design your own.

The best way to get good product taste is to try a lot of products and be opinionated about them

I’ve always enjoyed grandiose visions and ambitious missions :) exciting!

Showing up every day keeps hurting like the first day.

Time to lock myself away to ship something.

Interesting take.

How do you test an AI agent when it never gives the same answer twice?

A practical two-layer approach, with lessons from Baselight AI and other agents...

This week writing my Sunday post has been a challenge.

Turns out I won’t have Internet signal for the rest of the week for… reasons (we rely on a single base station in my current location and they decided to switch it off).

Just.

It still amazes me how fast money can be moved around these days

One of the risks of using coding agents is feature bloat (as getting new features is virtually free).

I’ve been trying to write a deep, insightful, and thoughtful tweet for minutes because it’s been a while since I post, but I have nothing.

Now delivering the two messages as specified: We outsourced manufacturing to China.

Qwen3.6 is a beast!

They are trying to find a sustainable cost model. > adlrocha: The economics of AI will start hitting hard!

The economics of AI will start hitting hard!

"The goal of some parents is to work so hard that their kids and grandkids get to live a life that appears spoiled by the standards of previous generations."

I found a clear weakness on all of my posts.

These LLM architectures are beautiful.

Last week’s post about the Claude Code leak proved something crucial: the real engineering in AI isn't the model, it's the scaffolding around it.

What the agent loops of open-source projects tell us about the future of software engineering....

I am actually really energised for next week.

I am seriously considering to buy a few GPUs and build my LLM rig.

You’ve probably already read this from me, but my role model in business is… Michael Scott

Good engineers are lazy and they are constantly trying to automate their job and get them obsolete

I would of course be stranded by now somewhere in the UK if it wasn’t for @whereisjorge’s help.

Things that can only happen to me, but have already happened a few times in my life: planning a trip thinking that I am landing at Airport A but landing at Airport B, just to realise my mistake when I disconnect airplane mode in landing, but in time to im…

Reality hits you hard, bro!

AI is the perfect for corporation to remove some fat.

Ok, this made me laugh out loud!

I am already running locally, in case you were wondering!

Google recently proved they can break Bitcoin's cryptography with 20x fewer qubits than we thought, without actually showing the attack circuit.

I was discussing this exact thing with @PabloGrueso the other day.

I called it the QA agent in my AI factory pipeline, but Claude releases again a feature that I just merged a few days ago.

You probably won’t believe me, but I haven’t used Claude in the whole week.

Time to update my local model!

The Claude Code leak revealed something nobody expected: The model isn't the product.

A lot of the hiring in the next few months will involve builders reimagining the internal tooling of companies to be AI-ready.

Testing AI agents is hard, and it requires a three-layer approach: 1️⃣ Scaffolding unit tests, deterministic checks for your orchestration logic 2️⃣ Issue-tagged regression tests, real bugs that broke production 3️⃣ LLM-as-judge + task-based evals to meas…

Don't you ever feel like there are certain people you just have a weakness for?

It’s been years since I don’t hate Mondays.

I am wondering if I should leave for some other time the post that I had planned for next week and analyse the source code of Hermes agents.

Today I am finally diving into the massive Claude Code leak and what it actually means for the future of building with AI.

What the Claude Code leak reveals about how agent engineering actually works...