Two goals you can see
-
Zero code-read time
You review decisions, not lines. A person approves what the product does and how the system is split into parts. Everything below that is generated from the approved design and held to it by the build: compiled boundaries, an audit that fails the build on structural violations, schema-validated design documents, and a check of delivered code against the design.
-
Zero coordination overhead
Work is split into parallel tickets only where the design proves the pieces are independent. Each component owns its files and tables, so parallel agents can't collide, and overlapping work serializes automatically.
Built for what AI is good at
-
Bounded inputs
Each layer of the design reads only the public surface of the layers above it, never the whole project.
-
Grammars, not vibes
Every generated design document must validate against a schema before it's accepted.
-
Strict, fast feedback
Generated projects are Elixir, chosen for its compiler and tooling. The build catches whole classes of agent mistakes.
-
Memory outside the model
Every rule carries its reason, so an agent can't simplify a rule back into the bug it prevents.
Fixed guarantees, open process
Catapult's opinions live in bundles: plain declaration files that define which design documents exist, the prompts that write them, where the review gates sit and how work moves to production. Every project can fork and edit them, and upstream improvements merge in through git.
The guarantees are fixed: enforced structure, frozen interfaces, the audit. How you work is yours.
Design once, deliver the MVP first
The full architecture for your initial feature list is designed and approved up front.
Then the MVP ships as one first build, and everything after it ships feature by feature, reprioritized by real usage.
Features you stop wanting never get built; they only ever existed as design documents.
Where it stands
Pre-MVP · In active development
- Built: the design engine, reactive scheduling, schema validation, a document-review loop where declined reviews feed back into regeneration, and the library generated projects inherit.
- Deliberately not built with itself: a flatter, less opinionated sibling pipeline builds it, so a Catapult outage can't take down the thing that builds Catapult.
- Next: end-to-end delivery of an application from design to shipped code, and an evaluation harness to measure all of this.
Who I'm looking for
-
Design partners
Small teams or solo founders with a new product who want to ship something large with AI.
-
Elixir shops
Teams that build many new applications.
-
Investors and collaborators
Not raising a seed round; a small SAFE would keep Catapult full-time through an end-to-end demo.
About and contact
Andrew Vickory
CEO, StrutCo
- Email[email protected]
- GitHubgithub.com/SwaggerAllen
- Substackstrutco.substack.com
- LinkedInlinkedin.com/in/andrewvickory