What a couple of months and 427,515 lines of agentic coding gets you

In May, Rill Data let its product and commercial teams go, having run out of runway and not raised a funding round. Startups can be a tough business to be in. However, it was also a great opportunity to see what a solo product builder can get done with agentic coding.

Claude Code may be addictive but it’s hella productive as well.

These are some of the raw data behind Mock Machines. This is a real product, built in the couple of months since Rill gave notice to its commercial team and engineers. It’s a tool I could have used in countless commercial situations at Rill, Retool, Looker. Building customised solutions needs good source data or APIs. It’s not vibe-slop, either: a crazy fast simulation loop measured in nanoseconds required care and attention.

Product (the useful output)

1 shipped product powerful, flexible simulation engine for pitches and prototypes ~8 ns per entity per turn optimised simulation engine 1.2M rows/sec bulk generation, on disk 4 integration surfaces REST API, warehouse export, Python library, LSP server 39 modelled scenarios business processes, reinforcement learning (MDP), cellular automata, Turing machines 10 custom data animations

Coding work (if you think these numbers are important)

387 pull requests 1,172 Linear issues 427,515 lines of code excluding comments

Polyglot programming including:

HTML 172,753

Go 121,820

JS / TS / JSX 33,012

Shell Script 3,737

Python 2,145

Rust 2,054

Lua — nvim config 43

Plus various others e.g. Dockerfile, go.mod, CSS, some XML.

Breakdown of deliverables

AI Features

Agentic editor with “fix loop” and context manager
Scenario generator
Warehouse-schema-to-simulated-scenario generator
Streamlined “Try Mock Machines” feature for website
Document generator with templating and AI-generated free text fields
Simulation support for Partially Observed Markov Decision Processes

Product Features

Separate views for modellers and business users
Brandable landing pages with guest links for sharing
Scenario compiler with comprehensive validation rules
Simulation runtime
Datastore (versions 1 and 2, now using ECS-style columnar storage for speed)
Built-in IDE with auto-completion and instant update of simulator
Catalogs for SKUs and inventory simulation
Faker library built-in for data generation
Taxonomy system with probability distributions for realistic categoricals
Checkpoint & resume for large simulations
Flush-passive-records system for massive beyond-memory simulations
OpenAPI interactive docs
LSP server for scenario building in your local IDE
Open Semantic Interchange specification generator
DuckDB exporter
Event log processor for supporting data animations
per-account git and object storage sync
Support for Markdown, HTML, Typst/PDF docs as part of simulated scenario
Rust/WASM based “docforge” editor

Website

Overall site structure and design
“Try Mock Machines” front page CTA feature
Shader generation for visual interest
Waitlist form & storage

Documentation

Static Astro Starlight site for users
Interactive Astro Starlight site for admins. Includes SolidJS islands connected to GitHub, Linear, and the production deployment

Development assists

Automated research and mockup process (sketch 24 mockups before building 1 product)