Agentic engineering
for software teams.

Your team has the coding agents. I make them pay off: the harness, the workflows, and the infrastructure underneath.

THE GAP

You bought the seats. The numbers didn't move.

  • Review is drowning. Agents open pull requests faster than your team can read them.
  • The token bill climbs every month, and nobody can say what it bought.
  • The agent keeps losing the plot on your codebase.
  • Your code can't leave the building, so half the tooling is off the table.
SERVICES

The harness, the workflows, the infrastructure.

One engineer across every layer an agent touches, from the instructions it reads to the machine it runs on.

Harness

The instructions, tools and guardrails your agents work inside.

  • Instructions your agents actually follow

    a shared CLAUDE.md / AGENTS.md architecture that stays lean as the codebase grows, instead of one bloated file nobody trusts.

  • Agents that can see your systems

    skills and MCP servers for your internal tools, docs and data, so agents stop guessing at context they can't reach.

  • Guardrails you can measure

    tool permissions, cost limits and an eval harness, so you know what an agent may do and whether a change made it better.

CALLS 01–03

Workflows

How your team and its agents ship together.

  • Review that keeps up

    feedback sensors, CI agents and PR-size rules, so agent output arrives reviewable instead of burying the people who merge it.

  • One way of working across the team

    a hands-on workshop on your own codebase, then 30 days of office hours while the habits set.

  • A harness that stays current

    models change monthly and harness assumptions go stale; a fractional lead keeps yours tuned, measured and re-evaluated.

CALLS 04–06

Infrastructure

Where your agents run, and what they cost.

  • Local, hybrid or hosted, by workload

    open-weight models on your own hardware for routine work, frontier APIs for the hard parts, routed by task.

  • Capacity math before the purchase order

    how many concurrent agents a box really serves, worked out from KV-cache headroom rather than the biggest model that fits.

  • Handed over, not hosted

    spec, build and a tested runbook your team owns: the method from my own reference build, documented in public.

CALLS 07–09

Engagements

Start small. Keep the fixes.

Every engagement is delivered by me directly, with no handoffs and no bench, for a few clients at a time.

Free check-up

30 minutes, one real fix.

Walk me through your setup and where it hurts. You leave with at least one concrete change to make this week, whether or not we work together.

Agent Workflow Review

Two weeks. Fixed fee. You keep the fixes.

I read your repos, agent setup, session logs and review history, talk to the team, and hand back a ranked findings memo, a baseline scorecard and a pull request with the top fixes already made.

How the review works →

Build-out & enablement

Three to six weeks.

The review's roadmap, implemented: instruction architecture, skills and MCP servers, review and CI wiring, guardrails and evals, plus a team workshop and 30 days of office hours.

Fractional lead

Part-time, sized to your needs.

Ongoing ownership of your agent setup as models and tools change: monthly measurement, tool evaluations, and a harness that doesn't go stale.

Insights

The method, in writing.

How I work, shown on my own systems: the builds, the failures and the fixes, written up in full.

AUGUST 16, 2026

The 128 GB box that replaced my API key

A Bosgame M5 with 128 GB of unified memory, turned into a headless Fedora server that runs 70B–120B-class models and drives a fully local, open-source coding agent — every decision, every gotcha.

Read →

All essays →Subscribe on Protocol Grade →

TRACK RECORD

Eight years under real stakes.

Before agents, I spent eight years at Sky (formerly MakerDAO), rising to lead the team behind its flagship app. Every claim below links to something you can open.

The numbers these frontends carry.

DATA AS OF 2026-10-30 18:27 UTC
$17.71B
Sky ecosystem TVL, per the protocol's official analytics
$4.49B
Held in the Savings module, fronted by a widget built end to end
810,085
Wallets across the Sky ecosystem, connected through frontends the principal built

PUBLIC, SOURCED, DATED — PER THE PROTOCOL'S OFFICIAL ANALYTICS · FIGURES REFRESH AT BUILD TIME

Sky.money — the flagship UI

The interface of the $17.9B Sky ecosystem — team-led from the ground-up redesign through today's five-network app.

● app.sky.money↗app.sky.money — Balances view of the live Sky.money application
FIG. 01 — CAPTURED 2026-07-07 · LIVE PRODUCT
ROLE · TEAM LEADNETWORKS · 5REPO · JETSTREAMGG/TARMAC ↗

An LLM assistant, live inside the product

Team-led delivery of the production frontend for Sky's chatbot, plus the documentation pipeline that generated its training data. Shipped to non-U.S. markets.

app.sky.money · non-U.S.↗The chat assistant panel inside Sky.money answering 'What can I do with USDS?' with save, swap, borrow and transfer options, plus action buttons and rating controls
FIG. 03 — CHAT ASSISTANT · PRODUCT UI · 2026-10-30

Black Thursday Emergency Operations

2020-03-12 — DEPLOYED OVERNIGHTOn the day the market broke, opportunistic bots were winning liquidation auctions for near-zero bids, threatening bad debt in the system. An emergency auctions dashboard, shipped overnight, put honest keepers back in the fight.

About

Phil Bain

Phil Bain runs Basswood Creative, an agentic-engineering practice for software teams, from Colorado. For the last two years he has built software almost entirely through coding agents, on frontier models and on his own local inference hardware, and he writes up the builds and the failure modes in public. Before that he spent eight years at Sky (formerly MakerDAO, 2018–2026), rising from developer to co-founder and team lead of Jetstream, the independent studio contracted as the protocol's primary UI team, and spoke at the Devconnect 2025 App Town Hall in Buenos Aires. He holds an M.A. in Emergent Digital Practices from the University of Denver.

Full resume → · LinkedIn · More of the person at philbain.com.

Are your coding agents paying for themselves yet?

Book a free 30-minute check-up