AI-assisted engineering
AI that moves fast — and still works.
I build AI into a modern SDLC: opencode, Cursor, MCP and a skill library for planning, testing and review. Every change is test-driven, reviewed and verified before it ships — AI sets the pace, fundamentals set the bar.
How I work
A skill per job. The tools below are the working set, not a shopping list.
Superpowers
Structured planning: brainstorm, written plan, test-driven implementation, verify before any completion claim.
PAAD
Agentic review skills: architecture, security and code review before anything merges.
Ponytail
Minimal solutions first: simplest thing that works, no unrequested abstractions.
TDD
Tests before implementation. Red, green, refactor: on every feature, AI-written or not.
OmniRoute
Cost-aware model routing: the right model for the task, not the default one.
opencode + Cursor
The coding agents I drive this workflow with.
MCP
Connects those agents to real systems and data, through a tested layer.
De-risking AI code
- AI accelerates; fundamentals decide. Every AI-assisted change passes the same gate: tests, review, verification.
- Spikes stay spikes: nothing from a throwaway build ships without the full process.
- Verification before completion: evidence before any 'done' claim: fresh build, fresh checks.
- Reviews run on every branch; agentic review catches what speed misses.
Case study · 2026
Wrapping a REST API with MCP to post vacancies everywhere
A global recruitment platform wanted AI agents to post and manage vacancies across multiple destination job boards, without rewriting the existing REST API.
The problem
The platform's existing REST API had no AI interface. Agents that could post and manage vacancies on multiple destination job boards would save hours, but only if they could act on the real system, safely, and be proven fast.
The solution
An MCP layer wrapped the existing REST API as-is, exposing its actions as tools an agent can call. Connected to an existing LLM, agents could create, update and manage vacancy postings across multiple destination job boards, proved end to end quickly, then hardened.
How it was built
- Superpowers: brainstorm the approach, write the plan, TDD the implementation, verify before claiming done
- opencode and Cursor drove the coding; PAAD skills reviewed the work
- Ponytail kept the layer minimal: a thin wrapper, not a rewrite
- OmniRoute routed model calls by task, capability and cost
- Unit and integration tests around every tool; defects fixed, then locked in as regression tests
Validation
Proved in 2026: an agent posting and managing vacancies on multiple destination boards, end to end, against the live API through the MCP layer.
Best for
Teams with an existing REST API who want AI agents acting on it, safely, testably, without a rewrite.
Not ideal for
Teams wanting a big-bang platform replacement, or with no appetite for tests.
Want this on your project?
AI speed with TDD, review and verification behind it. If you want an MCP layer, an agent workflow or a safe AI pilot, that is exactly what I build.
Talk about your project