Operations systems
The layer a services business runs on: margin analytics tied out to the penny against finance’s own exports, revenue-recognition tooling, billing portals, master-data hubs with deploy-time lint.
01 / AI-POWERED SYSTEMS · AUTOMATION · BUILT UNDER GUARDRAILS
The harness is
the product.
Most of the code and assets here were generated by AI — on purpose, and that’s the point. AI isn’t handed the project; it’s handed a system to build inside: published rules, gated deploys, metered spend, reconciled numbers. The 95% that’s generation gets trustworthy enough to ship because the 5% that’s judgment built the guardrails first.
THE FLEET, COUNTED
· by the pipeline, not by hand · snapshot 2026-08-16
10,871commits, all-time
132repositories
~665Klines of code, active fleet
27apps on one gated deploy
every commit is a deploy · every deploy passes a secret-scan + smoke gate
the named version: LinkedIn →
The bet this page is making
The industry keeps asking whether AI can replace the builder. Wrong question. The right one: what does one builder ship when AI does the generation and the builder does the governing? Everything below is the answer run as a live experiment — one person, a fleet of production systems, because thousands of lines of scaffolding exist for AI to build from, not for AI to reinvent.
That inversion is the whole practice. Rules that load before any code is written. A deploy command that refuses secrets and failed smoke tests. A spend meter that reconciles nightly against the provider’s own bill. Give AI a blank page and you get a demo. Give it a harness and you get systems that run for years.
You don’t hand AI the project. You hand it the system the project must survive.
ONE WORLDVIEW, THREE DIRECTIONS
AI as workforce, utility, and witness.
Workforce — creation under guardrails
Code, images, video, voice, music — generated in batch by self-healing pipelines with grading loops and zero per-output hand-touching. When a run fails, the fix lands in the harness, never in the artifact.
Utility — consumption under meters
Every AI call in the fleet lands as an attributed row: which app, which feature, which provider, estimated cost. Multi-provider routing with health probes and fallback; daily caps; a killswitch; the expensive model gated behind explicit sign-off, never a default.
Witness — judgment on the record
When AI helps judge a codebase, the judgment ships with its receipts: deterministic fact harnesses, file-and-line citations, arithmetic in code. A claim that can’t be cited gets deleted, not hedged.
The connective rule
Same standard in all three directions: nothing is trusted because the model said it. Facts are derived, spend is reconciled, outputs are graded. AI is leverage; the guardrails are why the leverage holds.
Where the work lives
The active fleet spans domains most consultancies would call separate practices. Counted from git — code files only, authored work only, client-assessment snapshots excluded.
AUTHORED CODE BY DOMAIN
· thousands of lines · derived from the repos, not estimated
25 production systems · one deploy pipeline · one database pattern · one secrets hub
Breadth isn’t the pitch. The same guardrails holding across all of it is.
What I build
AI creation pipelines
Generative production at batch scale — image, video, voice, music, code — with grading loops, character consistency, and GPU orchestration. Self-healing by rule: fixes land in the harness.
AI governance & cost control
Multi-provider routing, per-app daily caps, full call attribution, a paid-API killswitch, and nightly reconciliation against the provider’s own cost report. The boring parts that make AI deployable.
Technical diligence
An evidence-cited read of a codebase on a deal clock — 55 published criteria, deterministic fact harness, every claim carrying a file-and-line citation, findings graded by what they do to the price.
Integration & automation
CRM platforms, payments, workspace suites, chat bots, browser automation, API reverse-engineering from real traffic. The connective tissue between systems that were never meant to talk.
Platform engineering
The discipline that makes a fleet possible: one deploy command with secret-scan and smoke gates, centralized secrets, shared cost monitoring, canonical helpers used everywhere instead of rewritten anywhere.
Exhibit A — a live instrument
The clearest sample of the standard: a scored technical-diligence instrument, built and run against a real acquisition-grade codebase. Facts derived by a deterministic harness, judgment on the record, arithmetic in code, web + PDF from one findings file.
51/100
DEVELOPING
· sample assessment — 53 criteria, one cell each, scored 0–4
cite-or-deleteabsence protocoljudgment register
deal-consequence gradingfounder test
full method → principles
Standing proof
A PE-backed consulting platform runs its operations on systems built here: live margin dashboards tied out to the penny against finance’s own exports, revenue-recognition tooling, a client billing portal, and AI triage across sales and delivery — all in production, all still running. Happy to share the name and walk through it on a call; the sponsor’s own diligence team has been through the codebase.
Behind that sits two decades and change in enterprise software — a national wireless carrier, a Fortune-50 hardware maker, an integration platform, a data-analytics pioneer; director level and above, generative-AI leadership since 2023. Built a 100-person automation organization that took $150M out of OPEX in five years; stood up innovation teams that did $10M in resellable solutions from zero in two. The named version is on LinkedIn.
A deck expires the day it’s presented. Evidence compounds.
Engagement shapes
01 · Diagnostic Sprint
Two weeks inside your workflows. Out: where the leverage actually is, what to build first, and what not to build at all.
Fixed fee
02 · Build Engagement
The system itself — scoped, shipped, documented, handed over running. Your infra, your data, your keys.
Fixed-fee project
03 · Operating Retainer
The layer keeps evolving after it ships. We stay on the hook for it — measured on outcomes, not hours.
Monthly retainer
04 · Technical Diligence
An evidence-cited read of a codebase on a deal clock — scored against published criteria, every claim checkable, findings graded by what they do to the price.
Fixed fee · deal-clock friendly
These are the shapes it’s taken so far. Yours may be different — that’s usually the interesting call.
Ask anything
tillo consulting · pick a question — or type your own (novel ones get answered live)
>