Three ways to work with AI. One clear way to find yours.
Most AI problems get simpler the moment you can see their shape. Here’s how we frame yours — and the three ways we can help.
Two questions locate almost any AI problem — not “what is AI,” but where your specific piece of work actually sits.
What shape is the work? — three modes
AI does work in three modes, and the right one depends on the work in front of you, not on how far along you are.
Automation
a task you can define precisely once and run reliably many times: a document transform, a data extraction, a scripted multi-step pipeline. The craft is in the instruction; the value is leverage.
Augmentation
work you and AI do together, turn by turn, where the path isn't fixed in advance: drafting, exploring, pair-building. AI brings speed and range; you bring judgment and context.
Agency
work you want running on your behalf inside rules you set: a configured agent, a monitored routine. The craft moves from doing the task to governing the system that does it.
These are three problem-shapes, not three rungs. A mature team runs all three at once; a newer team sometimes needs an agent before it needs much automation. The question is which shape is your work — never how far up a ladder are you.
How ready are you to direct it? — four habits
In every mode, four habits keep human judgment in the loop. Where a team is strong or thin across them is the clearest signal of which door fits.
Delegation
knowing what to hand to AI and what to keep. Needs both domain expertise and a real read on what today's AI can and can't do.
Description
directing it well: what you want (the output), how it's approached (method), and how it should behave (tone, guardrails).
Discernment
judging what comes back: the output, the reasoning behind it, and the behaviour during the work.
Diligence
owning the collaboration: choosing tools thoughtfully, being honest about AI's role, and standing behind what ships. For regulated and enterprise work, this is the whole game.
Your work in one view
Each cell is a question we ask of your real work — a problem-locating lens, not a score, not a ladder.
| Delegation | Description | Discernment | Diligence | |
|---|---|---|---|---|
| Automation | What can be scripted? | What precise instruction? | Spot-check + test coverage | Audit trail + human-in-the-loop triggers |
| Augmentation | What stays human-led? | Turn-by-turn context | Evaluate mid-collaboration | Cite AI's role in the work |
| Agency | What policy governs the agent? | The agent's brief / guardrails | Monitor + alert on behaviour | Clear ownership when it acts alone |
Based on the AI Fluency framework by Rick Dakan, Joseph Feller, and Anthropic, licensed under CC BY-NC-SA 4.0. Adaptation's exposition; not affiliated with or endorsed by Anthropic.
Here’s how we’d point you
We put your real work in the grid and ask the questions that locate it. Where it lands tells us the door — and we’ll tell you straight which one we’d start with.
If your team can't yet confidently hand work to AI or direct it well — start with Educate. The gap is fluency; building an agent on top of a team that can't yet govern it just moves the problem. Get the capability first. Explore Educate →
If the work is clearly shaped — a defined task, a build-with, an agent to configure — and you want it built and handed over — that's Build. We scope it on one real workflow and prove it on your data. Explore Build →
If the burden is keeping a system honest as it runs — monitoring it, judging it, standing behind it — and you want the outcome without the operations — that's Run. Explore Run →
Most situations have a clear first door; some want two. We recommend the one that fits where you actually are — not the biggest one.
The model layer is your call either way. Already committed to a provider? We build on yours. Don’t want to manage one? We host and run it for you.