The agentic spectrum: how much of your market will be a machine?
Most of the conversation about AI agents assumes a single future arriving for everyone at the same time. I don’t buy that. The coffee shop on my corner and a cloud infrastructure company aren’t standing in the same weather, and they never will be. One sells to people who choose with their senses, the other increasingly sells to software that chooses on evidence. Treating those two as the same agentic problem is how you either panic too early or prepare too late.
Here’s the way I’ve started thinking about it with clients, and I want to test it in the open.
What the agentic spectrum measures
The agentic spectrum is the share of your clientele that will be an agent rather than a human. It isn’t a number you choose. It’s a measurement of where your buyers are heading. A brand low on the spectrum still sells to people making their own decisions. A brand high on the spectrum is already being evaluated, shortlisted and sometimes bought by machines acting on a person’s behalf.
It’s worth measuring because it tells you how hard you have to engineer your digital presence to be acted on, rather than merely read. A human forgives a thin website. They infer what you meant, they call to check, they give you the benefit of the doubt. An agent does none of that. When the buyer is a machine, your brand is either legible and corroborated, or it’s absent from the shortlist, and there’s nobody to phone.
What pushes a business up the spectrum
Four things move a business up. How comparable and commoditised the offer is, because agents thrive on like-for-like comparison. How machine-consumable the product is, since an API is native to an agent and a haircut is not. How routine the purchase is, because reordering and renewal delegate easily while once-in-a-decade decisions don’t. And how much the decision leans on human trust, emotion and physical presence, which pulls hard in the other direction.
Where I think businesses sit
These are my placements. The percentages are deliberately rough, because the point is the ordering and the reasoning, not a decimal.
| Type of business | Agent share | Why it sits there |
|---|---|---|
| Local coffee shop, restaurant, bar | ~5% | Chosen by people on mood, sense and place. An agent rarely picks your espresso. |
| Hairdresser, dentist, local trades | ~10% | Trust and physical presence dominate. Agents may find you, humans decide. |
| Boutique and experiential retail | ~15% | The purchase is sensory and discretionary, which resists delegation. |
| Estate agency, high-value considered services | ~20% | Big, emotional, human-led decisions. Agents assist the research, not the choice. |
| Hotels, travel, restaurant bookings | ~35% | Agents increasingly compare and book, though taste still intrudes on the final call. |
| B2B professional services and agencies | ~40% | Shortlisting goes agentic. The relationship and the close stay human, for now. |
| Consumer packaged goods, commodity e-commerce | ~50% | Routine reordering and price-led buying suit an agent almost perfectly. |
| Insurance, lending, commodity financial products | ~60% | Comparable, rules-based and repetitive, which is exactly what agents are built to sort. |
| Logistics, suppliers, procurement-led B2B | ~70% | Purchasing is already systematised, so the agent is a short step from the process. |
| Software and SaaS | ~85% | Agents evaluate, integrate and increasingly procure, often with little human in the loop. |
| Developer tools, APIs, cloud and data infrastructure | ~95% | The customer is frequently a machine already. The human is a distant approver. |
Rough placements, offered to be argued with, not cited as fact.
Why your position changes what you build
Your place on the spectrum sets your priority. Low down, you engineer for humans first and make sure the occasional agent isn’t confused. High up, you engineer for machine comprehension as the main event, because the machine is now the gatekeeper to most of your market. Get this the wrong way round and you either over-invest in agent-readiness for a market that still buys with its eyes, or you leave your brand illegible to the software that’s quietly deciding your future.
Argue with me
These placements are mine, and I’ve stuck my neck out on purpose. I’d rather be wrong in public and corrected than vaguely right in private. Tell me where you sit, and tell me where you think I’ve misjudged your industry. The companion to this piece is the delegation boundary, which asks a different question altogether: not how many of your buyers are machines, but how far each of those machines is trusted to go.