I’d love to know roughly how big it was in terms of hours of work. I’m particularly interested in real-world projects that were accepted by the client and paid for.
Evgueny lemasovs
Posted on
What’s the largest project you or any programmer you know has completed entirely with AI assistance?
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (4)
On the hype wave, when everyone around is talking about AI agents, security, the quality of code written by AI, and so on, I somehow don’t see stories about successful projects in which all (!) "foreach" were written by AI. What is the size of such projects? 100 hours, 1,000 hours of work?
77,000+ lines of TypeScript, all built with Claude Code over the past several months. Zero lines written from scratch by hand.
The project is TxDesk, a multi-tenant B2B SaaS platform: AI support agents for DeFi protocols that decode blockchain transactions in plain English. Full stack including API, embeddable widget (Preact + Rollup), website, shared package, Stripe billing, WalletConnect wallet verification, Discord and Telegram bots. 46 blockchain integrations, 30+ AI tool functions, 1,700 tests passing.
It took months but not because the AI was slow. It was because I kept shipping new features and integrations back to back. The actual coding velocity with Claude Code is insane, things that would take a team weeks ship in hours. I run 2-3 terminals in parallel on different parts of the codebase.
The trick that makes it work at scale: decision documentation. Each terminal has no memory of the others, so I write markdown docs capturing every architectural decision and load them into each session. Without that the AI happily undoes work from another terminal.
Now I've shifted to using AI to help with marketing and distribution too. The building was the easy part. Getting customers is where I'm stuck.
Live in production at txdesk.io.
65,000+ lines of Rust across 16 crates. A Layer 1
blockchain with BFT consensus, built from scratch
over about 30 weeks.
I designed all the architecture, made every decision
about consensus, state management, transaction types,
networking, and security. Claude Code wrote most of
the implementation under my direction.
The key thing people miss about AI-assisted coding at
this scale is that the AI doesn't make architecture
decisions for you. It can write a function, but it
can't decide whether your consensus engine should use
3-chain commit or 2-chain commit. It can't decide
your state tree structure or your transaction encoding
format. Those decisions are still 100% human.
The ratio is roughly: 90% of the lines were written
by AI, but 90% of the time was spent on decisions
the AI couldn't make.
It's open source if anyone wants to look at what
AI-assisted code at this scale actually looks like:
github.com/0x-devc/NOVAI-node
I wonder about the long term as well. Are they coming back for more projects? Totally happy a few months later?