Intelligence is being centralised. We're building the alternative. Co-founder @Nano_Collective | Nanocoder · sovereign AI infrastructure | Ava Technologies.
England, United Kingdom
Joined October 2018
- Tweets2.3K
- Following582
- Followers1.3K
- Likes74.9K
Ben Parry ⚡ Sovereign AI retweeted
Building something I’ve wanted for a while.
AI-powered code reviews like CodeRabbit are extremely useful but they can be out of reach for small developers and small teams. If you need a decent amount of reviews spending $48+ can get expensive quickly. So we’re building an AI-powered code review tool for GitHub PRs that’s super cheap, super fast, and accurate.
We’ve been working on it behind the scenes for a while. Still polishing a few things, but we’ll be launching soon.
Add your project ⬇️
Nanolist is live.
A community-curated directory for discovering AI tools, with a focus on open-source, local-first, and privacy-respecting software.
Submit yours → list.nanocollective.org
We’ve rebuilt the Ava Technologies website so it actually reflects what we do: helping organisations adopt AI properly, not just experiment with it. Discovery, policy, training, tooling, rollout.
avatechnologies.org
The AI job losses won’t come from companies automating their people out. They’ll come from companies that never hired those people, outcompeting the ones that did.
Ben Parry ⚡ Sovereign AI retweeted
We’re planning to upgrade prompt-scrub, so we opened a bunch of new issues covering features, security, detectors, CLI improvements, and workflow integrations.
Plenty of work for contributors 🚀
"The agents got good enough that we removed the human" sounds like progress.
Taken to its end, it is the same centralised automation logic that removes people from control entirely.
The answer is not humans checking everything. It is humans owning the rules, on infrastructure they control.
An AI you are asked to trust should live somewhere structurally incapable of turning against you.
Not inside a company that promises it will not.
A promise is only as durable as the incentives behind it. And incentives change the moment the business needs them to.
Compliance is where this gets real.
A lot of what we call human judgement in compliance is just consistent rule application. Machines are better at consistency than people are.
What machines cannot do is decide what the rules should be. That is where the human belongs.
The naive version of AI automation: let the machine do the whole thing and hope it got it right.
That is how you get confident nonsense published under your name, or a contract nobody read.
The failure is not that AI was involved. It is that no judgement was placed where the outcome mattered.
Good agent software does not try to remove the human.
It removes the drudgery that was burying the human, so their judgement lands where it counts.
The machine does the labour. The rules do the routine checking. The person does the part that is genuinely theirs.