AgiRook
Who is on the job. You choose which agent can run. That association holds.
Staff Engineer, AI Systems
I design agent systems, and I use them to ship real software.
Staff-level work. Permanent or contract.
The path
17 working prototypes
440+ tasks orchestrated
~80 min idea to prototype
One system. Two repos.
ContextForge
The project index and the written-down memory. Decisions persist. The job does not start over.
PocketBridge
The workbench, if you want one. Not required.
How one person ships this much.
A long coding job stays the same job. Who is on it, the index, and the written-down memory hold. The dumps get smaller as a result.
You already have a coding agent. AgiRook says who is on the job. ContextForge keeps the project index and the memory cards. A long session stays consistent because the dumps get smaller and the decisions stay written down. Prune is what falls out of that, not the product. Two repos. PocketBridge is optional.
You already have a coding agent. AgiRook decides what can run. ContextForge can be missing and the run still happens. What ships is working software. PocketBridge is optional.
Install is not the same as on. On is not the same as spend. The rules are files. Every run leaves a record.
A single run is one command. A job is a schedule, not a chat.
If ContextForge is missing, the run still happens.
PocketBridge can start a run and show the record. You do not need it.
AgiRook
Who is on the job. You choose which agent can run. That association holds.
ContextForge
The project index and the written-down memory. Decisions persist. The job does not start over.
PocketBridge
The workbench, if you want one. Not required.
AgiRook and ContextForge are in incubation. The last generation is on the architecture page.
Everything else, with an honest status on each.
Status
Products
Mango-QR
A browser studio for QR and AR campaigns. Edit in the studio, the phone updates in about two seconds, every scan feeds a live picture of engagement.
React · three.js working demoPowerOps
A back-office desk for energy trading. Plain-English requests, seven hard-coded rules, a log that cannot be quietly edited. Demo on labeled fictional data.
TypeScript · Cloudflare live demo · prototypeSniffSafe
Point a phone at something and find out if it is safe for a pet. The model only names the item. The verdict comes from a cited database. A test makes it impossible to return "safe" without a source.
Cloudflare · Expo live API · prototypeJira Insights
Ask plain-English questions about Jira delivery data and get answers backed by validated SQL.
TypeScript · Cloudflare Workers · D1 built · pre-deployHeadlands & Co
A live marketing site and lead-scoring pipeline. Parked.
Node · Cloudflare Pages · Python parkedThe Factory
You answer nine questions. About 80 minutes later you have a running app.
Claude Code · Cloudflare · Expo proven · v1.1tip-split
A no-account restaurant bill splitter — the app The Factory built end to end to prove itself.
Cloudflare Workers deliveredShipwright
An idea-to-build completion platform. In progress, private.
TypeScript · Cloudflare in progress · private
Agentic systems & tooling
Ostia-Aevum
A server that hands AI agents pre-validated building blocks instead of letting them guess.
TypeScript · Cloudflare Workers · MCP livelathe
A model-free pipeline that turns real, cited pain into scored project ideas.
Python publicmcp-secrets
Encrypted credential vault and transparent launcher for MCP servers. Keeps API keys out of plaintext client config.
TypeScript privategavel
The pre-publish check for open-source repos: secrets, structure, and a README that works on a fresh install.
Python publiccredscrub
Zero-dependency secret detection and redaction for text streams.
TypeScript privatememdash
A dashboard for the agent memory: a map of what the system knows and a queue for what has gone stale.
Node · node:sqlite in useaseprite-mcp
Lets an AI agent draw and edit pixel-art sprites by driving the Aseprite editor.
TypeScript · Node · MCP working toolCouncil
A panel of AI personas that argue a decision properly and report where they agree and where they still disagree.
Claude Code · Obsidian in useDelivery Highway
A shipping pipeline with hard gates, so nothing reaches users before it has earned each stage.
PowerShell · Pester phase 1 liveThinking-Tools
Three reasoning tools that switch on by context: brainstorm wide, decide on a ladder, ramp up fast.
Claude Code in use
Games & interactive
Come to Life
A game in Unreal Engine 5.7 with an AI-generated, painted-on-3D art style.
Unreal Engine 5.7 · C++ greyboxruneforge-frontier
A browser 3D city-builder you can walk through in first or third person.
Three.js · Vite early prototype
Systems programming
UniversalControler
Turns a game controller into full Windows control: pointer, keyboard, and text entry.
Rust · Direct2D WIP
Explorations
iOCT
A design exploration for a Telegram-native operations and analytics tool.
Python · Telegram conceptRules the system runs on.
The rules that keep quality from sliding while one person ships a lot. Each one started as a real mistake, and each one is now enforced by a script or a review step.
Frame before you build
Big or hard-to-undo moves get one pause first: name the goal, sketch two approaches, keep the plan visible.
Verify in stages
Done gets proven in stages — automated checks, then evidence, then a human. Look-and-feel calls are always human.
Floor and ceiling
Every quality bar pairs a checklist for the minimum with a pinned example of great, so "good enough" cannot drift.
Prove it, don't just build it
Nothing counts as done until it clears one real task end to end. Passing tests is not proof.
The ratchet
Standards only move up. Every lesson raises the bar and gets written down, so no shortcut can quietly lower it.
No hype
Nothing gets called good until it has been seen working. State what happened and point to the proof.
I'm Jared Frieden. I design agent systems, and I use them to ship real software. Staff-level work: the system that decides what can run, the tool that keeps a run honest, and the judgment about what ships.
I run a one-person practice that works like a small team. It attaches to the coding agent I already use. I stay the one making the decisions.
Brief in, working product out.