Jacob @jakecodes

Building Kai https://t.co/7FFToZ0W6Q howth.run Philadelphia Joined December 2015

Tweets

2K
Followers

3K
Following

817
Likes

7K

Jacob @jakecodes

21 hours ago

kaicontext.com

0 0 0 29 0

View Details

Kai is code review with verification. Before it posts a finding, it opens the code and checks whether the problem is real. If it can't confirm it, you never see it. On a 1,027-line pull request it raised 11 possible problems, opened all 28 files, and ruled out every one. What it posted was a clean review and the 2 decisions that need a person.

1 0 2 179 0

View Details

Jacob @jakecodes

6 days ago

kaicontext.com

0 0 0 43 1

View Details

Jacob @jakecodes

6 days ago

Kai is version control first. It's built in three layers. 1. Version control built on a semantic graph. The graph maps how your code connects, and it tracks how that map changes over time. 2. A harness that reads and writes code through that graph. 3. A code reviewer that uses both. The reviewer is the smallest part. The real power is in the versioned graph and the harness, and I think that graph is what the future of AI coding runs on.

2 0 3 208 1

View Details

Jacob @jakecodes

a week ago

github.com/FalkorDB/falko… kaicontext.com

0 0 1 37 0

View Details

Jacob @jakecodes

a week ago

In the next version of Kai, the reviewer works like a teammate. You can talk it through a finding. On this PR the author showed Kai the git history, and Kai said "Fair. Dropping the finding." You can also ask it to fix things. It makes the fix itself instead of handing it off to someone else.

1 1 2 193 0

View Details

Jacob @jakecodes

2 weeks ago

KaiContext: a semantic graph with version control built in. We build it from your last 50 commits, then review with small open models + custom MCP tools. Low cost, verified every step. Same code, same score, every time. kaicontext.com

1 0 4 172 0

View Details

Jacob @jakecodes

2 weeks ago

We’re grinding on Kai’s code-review benchmark score. With almost no effort we jumped a lot. Then we got stuck on the missing ~10 percentage points from the leader. Kai’s reviews were already strong - with bugs still in our reviewer we were scoring above CodeRabbit and Greptile. The leftover gap looks like tiny bugs on our side. Clearing those should put us over the top of the board, we hope. We’ll keep you posted.

0 0 1 137 0

View Details

Jacob @jakecodes

2 weeks ago

We wrote up how to read Martian’s bench → kaicontext.com/blog/how-to-re…

0 1 1 63 0

View Details

Jacob @jakecodes

2 weeks ago

🐛 Best benchmark for an AI code review tool: your own bug history. Grab 5–10 PRs that later needed a fix. Run each tool on the code before the fix. See who catches it.

2 1 1 180 0

View Details

Jacob @jakecodes

2 weeks ago

How it works → kaicontext.com

0 0 1 36 0

View Details

Jacob @jakecodes

2 weeks ago

In the next version of Kai… it flags bugs the diff never touched.

1 1 0 134 0

View Details

Jacob @jakecodes

2 weeks ago

🚞 Behind @kaicontextdev is a full version control system that lets the agent reason about your code using a graph over time and space. 🚄 When Kai reviews a pull request, Kai clones your repo and analyzes it very quickly. 🌳 It's backed by a semantic graph. 🧬 This lets Kai review code like I do, thinking of the whole codebase. 👷 I am solidifying this version control system, which will make review even more incredible than it already is.

0 0 2 151 0

View Details

Jacob @jakecodes

2 weeks ago

I'm hiring a full-stack engineer at Kai Context. We're building an AI code reviewer that checks every PR against a map of your whole codebase, not just the diff. Go and TypeScript, remote, real equity. DMs open, or email [email protected]

14 1 15 1K 1

View Details

Jacob @jakecodes

2 weeks ago

dots a lot of permissions to give @OpenAI

0 0 0 115 0

View Details

Sindre Sorhus @sindresorhus

2 weeks ago

I see this less as GitHub failing and more as open source evolving. The current workflow is increasingly absurd: someone uses AI to generate a PR, I use a better AI to review it, then their AI and my AI go back and forth until we converge. Add AI-generated spam on top, and the human is mostly just acting as a router between agents. I’d rather skip that entire loop. Imagine you use a project and want a feature. You go to the repo and describe what you want. The project’s AI already knows its conventions, preferences, history, rejected ideas, and boundaries. It can immediately tell you whether the idea is likely to fit and what shape it should take. If it makes sense, you press a button and the project’s chosen AI implements it, validates it, and presents the maintainer with the request and the code as one thing. No separate issue. No PR ping-pong. No two AIs talking through two humans. What AI still canot replace is taste: deciding what a project should be, what belongs in it, and where its boundaries are.

21 37 267 24K 55

View Details

Jacob @jakecodes

2 weeks ago

🔎 Found that the best way to avoid AI slop is to draw the website on a piece of lined paper. Then hand that to AI. Paper to result: @kaicontextdev website.

0 0 1 120 0

View Details

Jacob @jakecodes

2 weeks ago

I studied music and art. Doesn't it feel like the art of the world is just getting so damn boring?

0 0 0 110 0

View Details

Jacob @jakecodes

2 weeks ago

🤬I'm so annoyed that AI CANNOT write like a human, no matter what I try. I mean, not even close. I tried building a model JUST FOR WRITING. My dad was an English teacher and school principal, and he sat with me night after night, showing me how to write like a pro. I KNOW HOW TO WRITE.

0 0 0 109 1

View Details

Jacob @jakecodes

2 weeks ago

When you don't exercise, you don't feel good. When you use AI too much, you stop exercising your mind, and you don't feel good. So I'm redesigning @kaicontextdev homepage.... BY HAND!