Applied AI Operations · Open cohort

Twelve weeks to put AI to work on your own job — and prove it.

A hands-on course for one person who wants to get properly good at AI tooling. How agents actually work, what they cost, how to keep them inside the lines, and how to build a workflow that holds up on an ordinary Tuesday. You bring a real task from your job. You leave with it running, measured and written down.

Apply for a seat See the curriculum
Live online · cohorts of eight · 55 live hours · certificate on completion
Who this is for

You've used the tools. Now learn to operate them.

Most people have asked a chatbot to write something. Far fewer can hand an agent a piece of real work, predict what it will cost, check the result properly, and do it again next week with the same outcome. That gap is the course.

01

Results you can't repeat

It worked brilliantly once and then didn't. You can't say why, so you can't fix it — and you've quietly gone back to doing it by hand.

02

A bill you can't explain

Tokens, context windows, usage tiers. You know it costs money. You don't know which part of what you did cost it, or how to make it cost less.

03

You've been volunteered

Someone has asked you to “bring AI into the team”. You'd rather turn up with a working example and a set of rules than a list of tools.

04

You want evidence, not vibes

You'd like to be able to say what changed — minutes saved, errors caught, quality held — and have the numbers behind it.

How it runs

Small cohort, live instruction, your own work as the material

Nothing is pre-recorded or self-paced. Every session is live with an instructor, and every one of them ends with you applying what was taught to a task from your own job.

12wks

Two live sessions most weeks. Four to five hours of your time, plus whatever you put into your own build between them.

8seats

Capped so every build gets looked at properly and nobody can hide at the back.

55hrs

Twenty two-hour modules, four three-hour build labs, and three private hours with an instructor. All live.

31:1s

Booked when you need them, usually before a lab. Your build, your blockers, one instructor.

What you'll learn

Five blocks, twenty modules, one build that runs through all of it

Each module is two hours: ninety minutes of teaching, then thirty applying it to your own task. The material generalises on purpose — constraining what an agent can see is the same skill whether you work in a code repository or a folder of client contracts.

Block 01

Foundations

10 hours · Weeks 1–3

Why agents behave the way they do, and how to get the same result twice.

  1. How agentic systems actually work

    The loop an agent runs, what it can and can't see, and why it fails the way it does.

  2. The tool landscape and what it costs

    Compare the platforms on how they execute work and what they charge, and justify a choice.

  3. Context: the variable that decides quality

    Assemble and constrain what a task can see, and explain why the output changed.

  4. Instruction design and reproducibility

    Write instructions that produce the same result twice.

  5. Verifying work beyond your expertise

    Check output you aren't qualified to judge directly.

Block 02

Cost and control

8 hours · Weeks 4–6

Predict the bill before you run the job, and keep the agent inside the lines.

  1. Token economics

    Why one instruction can burn millions of tokens, and how to estimate the cost first.

  2. Reading and attributing a usage bill

    Trace spend to workflows and people, and spot the anomaly.

  3. Guardrails and configuration

    Write configuration that constrains an agent, and test that it holds.

  4. Context pruning and exclusion rules

    Keep binaries, generated files and noise out of the context window.

Block 03

Applied workflows

10 hours · Weeks 6–8

Five patterns that cover most knowledge work. You go deep on the one closest to your job.

  1. Research and long-form drafting

    A repeatable loop for research, drafting and revision of reports and proposals.

  2. Structured extraction

    Turn messy sources into structured, checkable data and a summary you can stand behind.

  3. Design and media production pipelines

    Run a pipeline with a quality gate at every stage.

  4. Agent-assisted software delivery

    Take a change from specification through generation, diff review and merge.

  5. Automation across existing tools

    Connect the tools you already use so a routine task needs less of you.

Block 04

Trust and governance

6 hours · Weeks 9–10

What may leave the building, who can touch what, and the record that survives scrutiny.

  1. Data classification

    Decide what can and can't be sent to a third-party model, and apply the rule.

  2. Security and access

    Credentials, permissions and third-party access in AI-assisted work.

  3. Disclosure, provenance and record-keeping

    A disclosure standard, and a record that holds up when someone asks.

Block 05

Making it stick

6 hours · Weeks 11–12

Measure honestly, recover cleanly, and write the rules you'll actually follow.

  1. Measuring time saved and quality held

    Instrument a workflow and report what changed, without flattering it.

  2. Failure modes, rollback and recovery

    Contain and reverse an agent-caused error without losing the work around it.

  3. Writing your operating standards

    The rules you'll work to, and the review that keeps them current.

The build

Four labs. One real workflow from your job, from scope to showcase.

In the first lab you pick a recurring task you actually do — the weekly report, the intake triage, the contract review, the release notes. Everything after that is in service of making it run without you.

  1. Week 3

    Scope

    Pick a recurring task you actually do, define what good looks like, and take a baseline measurement. If it can't be measured, we'll help you pick one that can.

  2. Week 6

    First working version

    It runs end to end, however roughly. Ugly is fine. Working is the point.

  3. Week 9

    Harden

    Failure modes, guardrails and cost. Making it survive an ordinary Tuesday, not just a demonstration.

  4. Week 12

    Showcase

    Demonstrate it, have it peer-reviewed, and present the measured before-and-after.

Each lab is three hours with the whole cohort: a short demonstration, then small-group work on your own workflows while an instructor rotates between you. Your three one-to-one sessions sit between the labs, for the blockers a group setting can't reach.

What you leave with

Things you can show, not things you've heard about

  • A working workflow from your own job, running on your own accounts and files
  • A one-page runbook anyone on your team could follow
  • A measured before-and-after: minutes, errors, quality and cost
  • A written operating standard — your rules for what gets automated, checked and disclosed
  • A cost model you can defend to whoever pays the bill
  • A Larkwin Certificate of Completion in Applied AI Operations
  • Ninety days of office hours after the course, for when the real world turns up
How it's assessed

A short quiz closes each module. The first three labs are peer-reviewed in your breakout group, against the milestone you set yourself.

The final submission is your workflow, its runbook and the measured before-and-after. Both instructors mark it against a published rubric: correctness, cost discipline, verification and documentation.

Honest about the paper: the certificate is a provider credential, not an industry one. The portfolio is what you'll actually show people.

Who teaches it

Two instructors. One knows the machinery, the other knows how work actually gets organised.

You work directly with both, in every lab and in your one-to-ones. No teaching assistants, no hand-offs.

IH

Ian Howlett

Founding Partner · fourteen modules, labs two and three

Twenty years in technology leadership, most recently Vice President at an Ontario software engineering consultancy, where he designed and delivered AI enablement programs for enterprise clients. Google AI Professional; Anthropic AI Fluency. Teaches how agents work, cost and control, software delivery, security and failure modes.

linkedin.com/in/ihowl
TK

Toni Kay

Founding Partner · six modules, labs one and four

Twenty years inside global banking, healthcare and large-scale transformation, including HR leadership at HSBC and CIBC and an interim directorship at Grand River Hospital. Teaches the workflows most knowledge work runs on — research, drafting, extraction, automation — plus disclosure, measurement and operating standards.

linkedin.com/in/tonikay
Is it for you?

For people with a real task to hand off

  • You already have a recurring piece of your job you'd like off your plate, and the access to work on it.
  • You can give it four to five live hours a week for twelve weeks, plus build time.
  • You work in operations, finance, marketing, people, grants and bids, research, or engineering — the method is the same, the task is yours.
  • Not for you if you want a tour of the tools, a certificate for its own sake, or someone to build it for you. We teach and supervise. You do the work.
Getting it paid for

It's built to be employer-sponsored. Ask us for the one-page brief for your manager — what the course covers, what you'll come back with, and what it asks of your week.

Sending three or more? The same curriculum runs as a private program for teams, with the labs built around your own systems. It may qualify for employer training grants such as the Ontario Job Grant; we'll walk you through what that needs.

Questions

The things people ask before they apply

Do I need to be technical?

No. You need to be comfortable with the tools you use every day and willing to read an error message instead of closing it. One module covers software delivery; it's taught so a non-engineer can follow it and an engineer can take it further.

Which AI tools do you teach on?

The ones your work runs on. The course is vendor-neutral: we teach how these systems work and how to run them, across the major assistants and agent platforms, and the method carries over when the tools change. We don't sell any of the platforms we teach.

How much time will it take?

Four to five live hours a week for twelve weeks, plus two or three on your own build between sessions. Recordings are there for reference, but the course is built around turning up.

What if I miss a session?

Watch the recording, bring your questions to office hours, and expect your breakout group to notice you were gone. Miss a lab and we'll use one of your one-to-one sessions to catch the milestone up.

Is the certificate recognised?

It's a Larkwin credential, not an industry one, and we say so plainly. What travels is the work: a running workflow, a runbook and a measured result you can show anyone.

Can my employer pay for it?

Yes. We invoice the employer directly and send you a one-page brief for your manager. If they'd rather send a team, the same curriculum runs as a private program, which may qualify for employer training grants such as the Ontario Job Grant.

What do I need to take part?

A machine you have administrator rights on, a working camera and microphone, and access to the accounts and files your chosen task touches. Course materials and the tool credits used in the labs are included.

Want a seat in the next cohort?

Tell us about the task you'd build around. We'll come back within a day with a straight answer on fit, the next start date, and the brief for your manager if you need one.

Apply for a seat