What we've built.

Our own portfolio: SaaS apps, a mobile game, AI services, training content. With agents doing the building, throughput stopped being the constraint. Coordination did — which product gets attention, which decisions are waiting on a human, and how agents can work unsupervised without anyone losing control.

• Built a central hub every project reports into and agents take their work from: stage, health, roadmap, backlog, live agent activity, and a decision inbox that surfaces what needs a person.
• Made the hub the single source of truth for requirements, decisions, stories, and plans; repo files are generated from it and synced both ways.
• Dropped a protocol kit into every repo so agents pull cross-project context at session start and report telemetry, decisions, and blockers back.
• Wrote the autonomy contract as code: ordered verification gates including an independent AI reviewer and a human walk-through, runaway limits, a merge policy where only small low-risk changes auto-merge, and a rule that agents queue a question rather than guess.

The hub manages itself as the tenth product in its own portfolio. It is the discipline we bring to client work — autonomy with explicit human gates and telemetry you can audit — proven daily on real products before it is ever proposed to anyone else.

A community organization needed a mobile platform for the frontline staff it supports: guidance, tools, and language support in the field, on their phones. What existed was a lightweight requirements document and a throwaway web proof-of-concept. What they needed was production iOS and Android apps, an admin console their own staff could run, and real AI capability — on a small-organization budget.

• Built the delivery system first: the requirements document and proof-of-concept became a full planning corpus — product requirements, functional spec, data model, service contract, and architecture decision records — decomposed into stories with acceptance criteria and a dependency-ordered build plan.
• Ran every story through a purpose-built AI production line: branch, build, test, review, pull request — with quality enforced by the harness rather than by hope, and per-story acceptance verification.
• Shipped a mobile app from one codebase, live AI services including voice-to-voice translation with per-user cost caps, content moderation, and semantic search, plus a complete CMS for content, media, users, and analytics.

Designed from day one to be owned, not rented: a portable architecture the client can re-platform without rewriting the apps, operator documentation for their team, delivery into their own repository with an auditable decision log, and content that lives in the CMS so the organization updates the app without engineering help.

The client sells specialized technology into a highly fragmented market: veterinary practices, specialist clinics, industry associations, cooperatives, universities, and government extension services spread across North and South America. They needed a sales-ready map of that market, and they needed to own the process afterward rather than depend on a vendor for every new region.

• Stage one, territory lead generation: AI research agents swept each country region by region against a defined category framework, in the local language, reaching registries and sources that commercial data providers miss. Every lead landed in a single CRM-ready schema.
• Stage two, deep research and scoring: each qualified lead individually researched for services, technology adoption, reputation, and fit — then scored and graded with a plain-language rationale so the sales team could prioritize without re-reading the research.
• Identified the key people per organization — named decision-makers and influencers with verified contact channels.

The leads were half the deliverable. The other half was the system: prompt and instruction templates for both stages, the category and scoring frameworks that keep results consistent across countries and languages, output schemas that import straight into their CRM, and a documented expansion process from territory scoping through QA. Their team opens new territories without us.