Agentic AI for ops workflows · San Diego

Agents that cut the ops grind — with you still in control.

I wire AI-driven automation into the recurring operational work — triage, remediation drafts, dataset mitigation, reporting loops — with human approval on anything that leaves the system. Real tools and tickets, not demos.

Fixed prices. A one-week Health Check is $1,500.

Illustration of a pine marten hugging a round-bottom flask
How it works

From repetitive ops loops to supervised agents

We pick one painful workflow, connect the tools your team already uses, and ship an agent with evals and approval gates before we widen the scope.

Pick the workflow

We choose one high-volume ops loop with clear success criteria and a safe blast radius.

Wire tools and evals

I connect the agent to your tickets and systems, add evaluation cases, and require approval on outbound steps.

Operate and expand

You get runbooks, metrics on time saved, and a path to add the next workflow without starting over.

Packages

Clear scope, fixed prices

Not sure where to start? The Ops Workflow Audit gives you a clear, prioritized list either way.

Start here

Ops Workflow Audit

$1,500

1 week

  • Map of recurring ops loops and handoffs
  • Where agents help vs. where process should change
  • A prioritized build list you can act on
Talk it through
Ongoing

Fractional Agent Ops

$5,000 / month

1 day a week · or 2 days for $9,000/month

  • Improve and expand agents month to month
  • Evals, incident review and mentoring
  • A regular check-in with your lead
Talk it through

Model and cloud costs are separate and paid directly by you. I'll estimate them in the plan, and I favor simple, low-cost tools where they fit.

Who it's for

Teams drowning in repetitive operational work

If the same triage, drafting and handoffs burn hours every week — and you want automation without giving up control — a supervised ops agent is the right next step.

On-call and triage heavy teams

Alerts and tickets pile up; people copy-paste the same investigation steps.

Data and platform ops

Dataset fixes, pipeline mitigations and status updates follow a pattern you can encode.

Ops leads who need leverage

You want AI help without a bot sending mail or changing prod unsupervised.

Teams with tools already in place

Tickets, chat and repos exist — the missing piece is a safe agent in the loop.

  • Usually a business of about 10 to 200 people.
  • At least one ticketing or chat system the agent can use.
  • You want approval gates on anything that leaves the system.
Headshot of John Houghton
Why work with me

Hi, I'm John Houghton.

I've built and run data pipelines at large scale, including multi-terabyte training datasets for AI researchers at Meta (contract) and the move of Apple Maps reporting from Hive to Snowflake.

I build smaller systems the same careful way: tested, documented and easy to hand over. You can see how I work in my public San Diego open data pipeline on GitHub.

  • Senior data and AI engineer. Previously Meta (contract), Apple Maps, Kaiser Permanente and Intuit.
  • B.A. Economics, San Diego State University.
  • Based in San Diego. You work with me directly, from the first call to the handoff.
Questions

Common questions

Will the agent act without a person?

Not on anything outbound. Drafts, ticket updates that need a human, and external sends go through an approval gate. Read-only steps can run automatically once you trust them.

How is this different from AI engineering?

AI engineering often means product features and LLM apps. This offer is aimed at operational workflows — the internal loops that keep the business running — with the same craft: tools, evals and guardrails.

Which tools do you use?

Whatever you already run for tickets and chat, plus the model providers you approve. I keep the architecture simple and documented so your team can extend it.

Who owns the code and data?

You do. Everything lives in your accounts and your code repository from day one.

What does it cost to run?

Mostly model and hosting cost. I'll estimate it before we start and design so quiet hours stay cheap.

Next step

Let's pick one ops workflow

Book a free 15-minute call. Tell me the loop that burns the most hours. I'll tell you honestly whether an agent is the right fix.

Prefer email? john@houghtonlabs.com