Ignacio

ai consulting director by day Working on System 2 Thinking by night. Engineering AGI.

[Ignacio @ignaflops]

That "plan locked" moment...

[Ignacio @ignaflops]

@Oluwaphilemon1 Can´t wait for the @neuralink MCP

[Ignacio @ignaflops]

If you are also stuck in Austin with this bad #airquality Sulforaphane has shown in clinical trials to help detox some air pollutants. Thanks @foundmyfitness for the discovery.

https://t.co/Os9bHrTS2K

[Ignacio @ignaflops]

idk what to answer to this

[Ignacio @ignaflops]

People introducing Rick Rubin's music production approach to software engineering are not software engineers. A mistake in a song can be part of its charm in a production environment can be your data leaked or your wallet empty.

Don't vibecode my banking app please.

[Ignacio @ignaflops]

Every time the agent makes a mistake, engineer the environment so it can’t make that same mistake again.

That’s the definition of Harness Engineering Mitchell Hashimoto gave, and I think it names a shift a lot of people are already living through without quite calling it that.

Over the last year, the conversation around agents expanded from prompt engineering into context engineering, while another layer kept growing in parallel. If you’re building with coding agents, you’ve probably already run into it.

Prompt engineering is how we shape the model’s instructions. Context engineering is how we shape the information and state the model has access to so it can make good decisions. Harness engineering is how we craft and tune the environment around the model so the agent behaves reliably for a specific job, with the right tools, permissions, memory, workflows, verification loops, and guardrails.

That is where the leverage is moving.

The problem is no longer just getting a strong model to produce a good answer once. The real problem is getting an agent to operate over time without drifting, breaking things, wasting money, or confidently digging itself into a hole.

The teams getting the most out of coding agents are not the ones writing the cleverest prompts. They’re the ones building environments where the agent can recover, verify, constrain itself, and avoid repeating the same class of mistake twice.

In practice that means things like architectural constraints, externalized memory through feature lists, progress logs, sub-agents used as context firewalls, self-verification loops with tests and evals, and feedback loops tight enough for the agent to catch its own mistakes before a human has to.

The bottleneck in AI-native software development is increasingly not the model. It’s human attention.

Harness engineering is how you reduce that tax. It’s how you turn “sometimes the agent gets it right” into “the system gets more reliable over time.”

Because once agents are good enough to be useful, they are also good enough to create expensive nonsense at scale.

At that point, the advantage no longer comes mainly from model access. It comes from how well you shape the environment around the model.

If you’re building with coding agents and you’re not asking, “How do I make this mistake impossible next time?”, you’re probably missing where the biggest leverage actually is.

[Ignacio @ignaflops]

This is good

https://t.co/ifOL84W1gU

[Ignacio @ignaflops]

You can use claude-code-router to call your local LLMs from Claude Code.

#claudecodedown #claudedown

https://t.co/xqgPAyfCkj

[Thariq @trq212]

We just added /btw to Claude Code!

Use it to have side chain conversations while Claude is working.

[Ignacio @ignaflops]

https://t.co/alLIcA8CTQ

[Ignacio @ignaflops]

https://t.co/Aj5TpdZuWf

[Ignacio @ignaflops]

@elonmusk With less than $10M a critical thinking school can be created.

[Ignacio @ignaflops]

New ideas run on red bull

[Ignacio @ignaflops]

@TwitterSupport If you need to build some intelligent automation to avoid this kind of situations. My company has a lot of experience creating DevOps solutions focused on scalability, reliability, resilience, low maintenance and autonomy.

[Lex Fridman @lexfridman]

Twitter Files saga continues to reveal abuses of power.

It's unfortunate to see. But, again, transparency is good for everyone in the long run.

[Ignacio @ignaflops]

The universe is a machine with intelligence as its output. Big Bang creates particles, they interact creating atoms and then dust, gravity turns dust into stars and planets, planet cycles create organic life, evolution makes life intelligent, intelligence synthesize itself.

[Ignacio @ignaflops]

Instead of claiming for censorship, let’s improve our critical thinking. We’ll become smarter and the world will be a better place.

[Ignacio @ignaflops]

@elonmusk Use a cryptographic proof on owning a public key with a verifiable identity history record and you'll have it solved. Easy peasy engineering!