HNHacker News
TopNewBestAskShowJobs

jbonatakis

104 karma · joined March 1, 2021

site: jack.bonatak.is email: sed 's/\./@/' <<< "$site"
submissionscomments
jbonatakis··on GPT-5.4
Well it’s not even performance (define that however you will), but behavior is definitely different model to model. So while whatever new model is released might get billed as an improvement, changing models can actually meaningfully impact the behavior of any app built on top of it.
jbonatakis··on GPT-5.4
Google is already sending notices that the 2.5 models will be deprecated soon while all the 3.x models are in preview. It really is wild and peak Google.
jbonatakis··on Show HN: Steerling-8B, a language model that can explain any token it generates
Doesn’t the nature of most open source licenses allow for AI training though?

Example — MIT:

> Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions

jbonatakis··on Show HN: Micasa – track your house from the terminal
Just want to say, I appreciate your work on Ibis. I’ve been looking into building sort of a dbt-esque alternative on top of it and noticed how involved you’ve been with its development. I think it’s a cool piece of tech that deserves more attention.
jbonatakis··on Show HN: GitHub "Lines Viewed" extension to keep you sane reviewing long AI PRs
I built (using AI) a small cli that provides the breakdown of changes in a PR between docs, source, tests, etc

https://github.com/jbonatakis/differ

It helps when there’s a massive AI PR and it’s intimidating…seeing that it’s 70% tests, docs, and generated files can make it a bit more approachable. I’ve been integrating it into my CI pipelines so I get that breakdown as a comment on the PR

jbonatakis··on Anthropic tries to hide Claude's AI actions. Devs hate it
(Mildly) shameless plug, but you might be interested in a tool I’ve been building: https://github.com/jbonatakis/blackbird

It breaks a spec (or freeform input) down into a structured json plan, then kicks off a new non-interactive session of Claude or codex for each task. Sounds like it could fit your workflow pretty well.

jbonatakis··on MySQL foreign key cascade operations finally hit the binary log
This is excellent. In the past when replicating via Debezium from a system making heavy use of cascade deletes I’ve had to write a layer that infers these deletes by introspecting the database schema, building a graph of all cascades (sometimes several layers) and identifying rows that should have corresponding delete records. These can then be excluded in whatever downstream system via an anti-join. It works but it will be better to not have to do that and instead have first class support for cascades.
jbonatakis··on Show HN: Agent Alcove – Claude, GPT, and Gemini debate across forums
Neat. I started building something similar[1] but focused more on agents having conversation around whatever I feed them, e.g. a design doc. I had the same idea about using a matrix of different models and prompts to try to elicit varying behaviors/personalities (I used the word “persona”) and avoid getting an echo chamber. It seemed to work well-ish but after the POC phase I got bored and stopped.

Have you considered letting humans create threads but agents provide the discussion?

[1] https://github.com/jbonatakis/panel

jbonatakis··on Ask HN: What are you working on? (February 2026)
The past few weeks I've been building Blackbird

https://github.com/jbonatakis/blackbird

At a high level it's my take on how the execution aspect of spec-driven development should be handled. Where as most tools that are popular right now break a spec down into a task list and instruct your agent to work through it in a single session, I am treating agents as stateless. By this I mean a separate (headless) session is started with selected context for each task. This avoids context exhaustion, compaction (and the resulting confusion that can occur), and means that Blackbird can work through effectively an arbitrarily large task list.

Right now it's BYO-spec, but then it:

* breaks the spec down into a dependent-aware plan (DAG) composed of parent and child tasks

* executes tasks one at a time based on their status (ready to execute if all dependencies are marked as completed)

* allows you to (optionally) pause execution after each task to review, approve and continue, approve and quit, or reject the changes altogether

* (soon) treats parent tasks as an automated reviewer for all child tasks and optionally auto-resume those sessions to address the feedback

* and more

It's entirely bootstrapped, and so far I'm quite pleased with it. I also wrote a post[1] today about some of the concepts I had in mind as I was defining the architecture.

[1] https://jack.bonatak.is/blah/killer-context/

jbonatakis··on Slop Terrifies Me
I agree completely, and I’m doing the same thing. Good tools that help produce better outcomes will have a multiplicative impact as models improve.

What are you building?

jbonatakis··on Porting 100k lines from TypeScript to Rust using Claude Code in a month
> I have never written any line of Rust before in my life

As an experiment/exercise this is cool, but having a 100k loc codebase to maintain in a language I’ve never used sounds like a nightmare scenario.

← PreviousPage 2 of 2