The agent advantage melts

Remember when chat-with-your-data was the showstopper of every trade fair? Same feature. Same code. A different commercial value.

Today it's a line in a requirements table. Nobody asks for a demo of it and nobody pays extra for it. The team that built it still maintains it.

After this you'll be able to point at your roadmap and say which items still earn a premium, which are pure cost, and where you're leaving a gap open.

Noriaki Kano's model (1984) sorts product features by how they affect customer satisfaction. A delighter surprises — nobody misses it, but its presence wins the deal. A performance feature gets comparison-shopped: more is better, and the price is negotiated. A hygiene feature is taken for granted: its presence pleases no one, its absence disqualifies you. And reverse quality is a feature that annoys more the better you build it.

The forgotten half of the model: features drift downhill. Yesterday's delighter is today's performance feature and tomorrow's hygiene. Growing agent capability doesn't repeal this law — it accelerates it.

This isn't a forecast; it's a rerun. Remember when a mobile app was a competitive advantage? Spell check, cloud sync, dashboards, single sign-on — every one of them entered the market as a delighter and retired as hygiene. Summaries, chat-with-your-data and copilots are riding the same conveyor, just faster.

You can bill for a delighter. Performance gets negotiated. Hygiene earns nothing — it's the ticket price for a seat at the negotiating table. Your cost structure, however, doesn't know this: a hygiene feature costs just as much to build and maintain as a delighter does.

That leads to an uncomfortable piece of arithmetic: the faster agent capability grows, the larger the share of your codebase doing work nobody pays for separately. A head start isn't an asset. It's a lease, and the landlord is the pace of model development — a landlord who terminates without notice. Every AI feature on your roadmap has a half-life, and each model generation shortens it.

While a feature is a delighter, you can skip it — you lose a premium, not a customer. But once that same feature has drifted into hygiene, skipping it no longer saves money. It forces your customer to buy the missing piece from another vendor — and from that moment on, your customer is your competitor's customer too.

Competitors don't come in through the front door. They come in through the feature you didn't build, and they spread out from there. Some AI features are making exactly this journey right now: what looks like an optional extra today will shortly be a condition for your contract getting renewed at all.

Not every AI feature drifts downhill — some are born below the line. The assistant that nags on every screen and the agent that spams your customers are reverse quality: the better you execute them, the more they annoy. "We have AI too" is not a feature; it's a risk. Agents belong on the buying side — helping your customers buy, not harassing them. And your customers spot that difference faster than your roadmap does.

The Kano drift is one lens. This series applies a stack of them to the same underlying question: how competitive advantage behaves when the marginal cost of capability collapses. Agents are the current instance — not the whole story.

Coming up: where to build (a Wardley map of your sales function), what stays defensible (7 Powers when junior work is free), when metrics die (Goodhart at machine speed), and the agent that annoys (reverse quality, in full).