Junayed @junayed711

Day 1 of Accelerate Cambridge done. GateBolt is on Cohort 40 at Cambridge Judge. Git tells you what a coding agent changed. It doesn't tell you what it meant to change. So GateBolt captures the intent upfront, checks it against the diff, and hash-chains the record. Day 1 done. #buildinpublic

GateBolt is joining Accelerate Cambridge at Cambridge Judge Business School, Cohort 40. I started GateBolt after running into this problem myself. I build with coding agents every day, and git tells me what changed. Nothing tells me what the agent set out to do. GateBolt makes the agent declare its intent up front, checks that against what actually landed in the diff, scores the gap, and seals the result so it can't be edited after the fact. If you run agents on your own repos and one has quietly gone somewhere you didn't ask for, or you're the person who'll be asked to prove what your team's agents did, I want to talk. gatebolt.com

I build everything with coding agents now. At some point I realised I couldn't answer a simple question about my own repo: which agent changed this, and was it meant to? Git tells you what changed. Nothing tells you what the agent set out to do. Code review assumed the author could explain themselves. Agents broke that. So GateBolt makes the agent declare which files it will touch, then scores that against the real diff. One undeclared .env outranks twenty test files. Code is where it starts, not where it stops. Try it without an account — runs in your browser: gatebolt.com/playground Running agents on a real codebase? Tell me where the score gets it wrong.

keyboard_arrow_left Previous keyboard_arrow_right Next

This weekend I'm in Cambridge. Building GateBolt at the moment — it checks what AI coding agents actually changed against what they said they changed. If you're around Cambridge this weekend and want to grab a coffee, let me know. Always up for meeting other people building things. #buildinpublic

In the demo I log in with the CLI, run gatebolt init in my personal website repo, then open Claude Code and ask it to make a change. Setup is one command. After that nobody types anything — the hooks handle declaring and reconciling on their own.

I recorded a short demo of GateBolt — first time showing it actually running rather than as slides. As teams adopt AI coding agents, it's hard to know what an agent actually changed versus what it said it would. A diff tells you what happened, not what the agent set out to do — so a drive-by edit looks the same as a deliberate one. GateBolt closes that gap: - Declare — the agent records the files it plans to touch before editing. - Reconcile — GateBolt compares that against the real git diff and scores the drift, escalating on sensitive paths like .env or auth. - Record — every step is written to a tamper-evident, hash-chained audit ledger. The scoring engine is deterministic — no LLM — so the same diff always produces the same verdict. Still early, but it's live — you can use it today. Would welcome thoughts, especially from anyone running agents against real codebases. gatebolt.com to get started, or npm i -g @gatebolt/cli. #buildinpublic

Claude's down and turns out I've forgotten how to code myself 😅

Day 63 of building in public. Feeling much better. Day job filled most of today, then travelling to London this evening. Two demos lined up: GateBolt to a law firm tomorrow, and to a VC on Wednesday. First proper demos to people outside my own head. #buildinpublic

Day 62 of building in public. Fallen a bit ill, so today was a proper rest day — not much beyond resting up. No code, and that's fine. Health first, GateBolt can wait a day. #buildinpublic

Day 61 of building in public. Mostly head-down coding today — squashing bugs and improving GateBolt, plus working on the pitch deck. Want it in good shape in case next week's London demo goes ahead. Nothing sharpens the work like a real demo on the horizon. #buildinpublic

Day 60 of building in public. Day job took most of today, but had a really interesting meeting about GateBolt with a potential customer. Might be heading to London next week to demo it to them. Early days, but encouraging. #buildinpublic

crit is here if anyone's curious → github.com/tomasz-tomczyk…

Day 59 of building in public. Most of today was the day job, then a startup demo event in the evening. Came across crit — an open source tool for reviewing AI agent output in the browser rather than the terminal. Interesting one if you're working with agents. No GateBolt time today, but a good evening. #buildinpublic

Day 58 of building in public. Work took most of the day again, then in the evening I travelled down to London — Zed x Tessl demo night is tomorrow, and since I'm in Leeds, wanted to get down early. Not much progress on GateBolt today, but looking forward to seeing what other founders are building. #buildinpublic

@attacomsian Done - connected :)

Day 56 of building in public. Spent today fixing bugs on GateBolt. The unglamorous part — but a tool that flags what AI agents got wrong has to be right itself. Also: I'm in London Wed, Thu and Fri this week. Would love to meet other founders while I'm down — if you're around, let's grab a coffee. #buildinpublic

Day 57 of building in public. Day job took most of today. Snuck in a hour to squash more bugs in the GateBolt CLI — the more I test it, the more edge cases turn up. Chipping away. #buildinpublic

This week I'm in London — Wednesday, Thursday and Friday. Thursday evening I'll be at the "Zed x Tessl We Love Demos event". Looking forward to watching startups demo what they've been working on. If you're around London this week and want to grab a coffee, let me know. Always up for meeting other people building things. #buildinpublic

Link to the pitch desk in Google Drive: drive.google.com/file/d/1b83l3a…

Last week on Friday I pitched GateBolt in front of two industry experts as part of Cambridge Judge Business School's Ignite programme — 10 min pitch, 15 min Q&A. GateBolt is only ~10 days old — it's the fourth product I've built, and it grew straight out of the pain points from building Layzer, Zepee, and BanterBird first. Looking back, what I'd improve: being crystal clear on the ask. And I'd have loved the chance to actually demo it live instead of just walking through slides. Overall — an amazing, genuinely unforgettable experience. Grateful to Ignite for the room and the hard questions. #buildinpublic #indiehackers