Discovery session.
What to expect.
Two hours, your office or mine, no slides. We map your bottlenecks, build one working thing, and you keep everything we make.
I run discovery sessions because I refuse to quote work without seeing the inside of the business first. Generic AI consulting is how you end up with a six-figure invoice for a chatbot nobody uses. Specific consulting starts with looking.
Here's exactly what happens when you book one. No mystery, no theatrics. The whole point is that you walk in prepared and walk out with something tangible.
Map. Sort. Build.
In that order.
The agenda below is the default. We can flex it if you arrive with a specific bottleneck already in mind, but most clients benefit from working through all three phases at least once.
Phase 01 · ~30 min
Map the day
We walk through a typical week, app by app, click by click. The goal is to find the workflows that actually eat time, not the ones that feel like they do.
›
Map the day
We walk through a typical week, app by app, click by click. The goal is to find the workflows that actually eat time, not the ones that feel like they do.
I'll ask you to literally show me your screen. Open the apps you use. Talk me through the Monday-morning routine, the Friday-afternoon report, the thing your operations manager keeps complaining about.
I take notes on three things: repeated clicks, copy-paste between systems, and data that arrives in a slightly different shape every time. Each of those is a different kind of opportunity.
You don't need to prepare a deck. The messier and more honest the walkthrough, the better the output.
Phase 02 · ~30 min
Sort the work
Each bottleneck gets routed into one of three buckets: code it, agent it, or leave it alone. Most "AI projects" get this wrong on day one.
›
Sort the work
Each bottleneck gets routed into one of three buckets: code it, agent it, or leave it alone. Most "AI projects" get this wrong on day one.
If a job has uniform inputs and a uniform output (a daily CSV, a templated report, a known API call), it goes in the deterministic bucket. We solve it with a script. AI compute is overkill and probabilistic drift will eventually bite you.
If a job has messy, non-uniform input (résumés, support tickets, free-text emails, a folder of inconsistent invoices), it goes in the agentic bucket. That's where the model earns its keep.
And some things stay manual. If a workflow is genuinely better as a human task, judgement-heavy, high-stakes, low frequency, I'll tell you that. The honest answer matters more than the billable answer.
Phase 03 · ~60 min
Build one thing
We pick the smallest meaningful win and ship it before time's up. A working skill, a script, a starter agent, something you can demo to your team on Monday.
›
Build one thing
We pick the smallest meaningful win and ship it before time's up. A working skill, a script, a starter agent, something you can demo to your team on Monday.
This is the non-negotiable bit. You don't leave the room without a working artefact. The form depends on what we mapped, could be a Claude Code skill that does your weekly reconciliation, a Python script that parses your messy invoices, a starter agent wired to your Telegram.
It will not be polished. It will work. That's the point. Polish is a separate engagement; proof-of-life is what discovery is for.
You keep all the code, all the prompts, all the configs. No SaaS account I control, no key tied to me, nothing you'd lose if I disappeared.
What to bring
Less than you'd think
- · A laptop with the systems you use day-to-day already logged in.
- · One person who can answer "why do we do it this way?" with more than a shrug. Often the founder, sometimes the ops lead.
- · A real example of one repeating job. Last week's report, an actual customer email, a representative invoice. Not a sanitised demo version.
- · A list of your current SaaS subscriptions (and a rough monthly spend).
- · Any AI experiments you've already tried, even the ones that flopped. Especially those.
- · A vague problem statement. "We feel like our finance team is drowning" is enough to start with.
What you walk out with
Three things, every time
A skill, script, or starter agent that solves one real bottleneck. Demonstrably running on your hardware before I leave.
A short document, like this one I wrote for Canex, covering what we found, what we built, and a sequenced roadmap. Yours to share internally.
A frank read on whether you need ongoing help, a one-off build, or just an internal hire and a nudge in the right direction. Sometimes the answer is "you've got this." That's a fine outcome.
Sovereignty audit
Where is your data going?
Most businesses have no idea how much of their proprietary data is currently passing through American hyperscaler infrastructure every single day. We sketch the map. As I argued in post two, agents are a much leakier boundary than chat, and most modern AI deployments are quietly graduating from one to the other.
Hosted frontier models
Best capability per dollar. Right answer for most workflows. Trade-off: your prompts and any data they touch live, briefly or permanently, on someone else's machine.
Local open-weight stack
Your hardware, your network, your data never leaves the building. Capability is a notch behind frontier. Right answer when sensitivity beats raw output quality.
We decide which workflows belong on which side of that line, and the rationale gets written into the debrief. Most businesses end up running a mix. A small set of sensitive jobs locally, the rest on cheap API calls.
Three honest paths
No retainers, ever
You take it from here.
Plenty of clients leave the discovery session with enough direction to run the playbook themselves, especially if they already have a sharp internal hire. That's a win for me. Send me a postcard when it ships.
Project-based build.
We agree on a defined scope from the roadmap, usually two to six weeks, and I build it. Fixed fee, defined deliverables. When it ships, the engagement ends. Easy to start, easy to finish.
Ongoing pilot training.
If your bet is on training a junior internal AI pilot, usually the highest-leverage move, I work alongside them as a coach for an agreed window. They drive, I review, they level up fast. Then I leave.
Ready when you are.
Brisbane in person, anywhere on a video call. The first session is on me, I'd rather earn the next one with a working artefact than a slide deck.
↓ Start a Conversation