English · Nederlands
Too many AI launches I see are a copied feature walking into a market that is already spoken for. Some come from start-ups wrapping a model in an interface; others from established companies bolting a chatbot onto their product. In 2011 I wrote about Localicious, a local-search app from WhitePages. Reading it again in 2026, I see the same three mistakes, at much higher speed.
What I saw then
Localicious was a free Android app built on WhitePages’ business listings: browse your neighbourhood, find places, read tips from other platforms, soon get deals. In other words, a check-in app in a year when everyone already had one.
I predicted it would fail for three reasons. Its features were not unique: neighbourhoods, automatic check-in and 21 million listings were all available elsewhere, and easy to copy. It was a standalone app in a market already divided between Google, Facebook, foursquare and the deal sites, so building its own network would cost millions. And it ignored the company’s real strength: a sales force that sold advertising to local businesses over the phone all day.
My advice was to sell location-based advertising through that sales force, integrate with every platform instead of competing with them, and get advice from someone not selling a single platform.
What is the same now
Replace “check-in app” with “AI assistant” and the post needs few edits.
The copied feature. Many AI products are a good prompt, a clean interface and a model someone else owns. That works until the model provider or the market leader ships the same thing. Jasper, one of the best-known AI writing tools, saw its growth slow after ChatGPT arrived as a cheap alternative almost overnight. In 2023 it cut its internal valuation by about 20%, lowered its revenue forecast, replaced its CEO and refocused on marketing teams in larger companies (Maginative, citing The Information). The model provider was also the competitor.
The standalone app. Established companies make the WhitePages mistake in reverse. They have the customers, the data and the sales team, and then launch AI as a separate product with its own brand, its own login and its own acquisition budget. The customer now has one more tool to learn, next to the general-purpose assistant they already use.
The ignored strength. Look at what an incumbent can do when it plays to its strength. In 2023 Thomson Reuters paid $650 million for Casetext, maker of CoCounsel, an AI legal assistant powered by GPT-4 (Thomson Reuters). It did not try to out-build the start-ups. It bought the capability and put it in front of the legal customers it already served. That is the move I recommended to WhitePages in 2011: bring the new thing to customers through the channel you already own.
What is different
Three things have changed, and they cut both ways.
Building is cheap now. Localicious needed its own network effect, which took years and marketing millions. An AI feature takes weeks. That lowers the cost of entry, but it also means your competitor’s cost of copying you is close to zero. Distribution, not the build, is where the money goes.
The platforms compete directly. In 2011 Google was a platform you could build on and a competitor at the same time. Model providers today are the same, but they move faster: a capability that was a start-up’s product in January can be a free feature of a general assistant by summer.
And a narrow product can win. Unlike a generic check-in app, an AI product that owns a specific workflow, with the data and the audit trail that come with it, can be defended. The wrappers in trouble are the ones that own nothing but the interface.
What to do about it
- Run the one-release test. List your AI differentiators and mark every one that a model provider or market leader could ship in a single release. What is left is your real proposition.
- List your unfair assets before your features. Sales force, customer relationships, workflow data, system of record, partner network. Put AI on top of those, not next to them.
- Sell AI through the motion you already have. Add it as an option your existing sales team and account managers can offer at renewal, before you build a separate brand and funnel for it.
- Partner or buy before you rebuild. Integrate with the assistants and platforms your customers already use. If a capability is missing, a partnership or acquisition is often faster than a two-year build.
- Kill copied features early. If a launch fails the one-release test and has no unfair asset behind it, stop investing. Every quarter you spend on it is a quarter not spent on what only you can do.
Same mistake, faster clock
In 2011 I said there was no excuse for repeating old mistakes after two years of watching the market. In AI we have now had several years. The winners will not be the companies with the most AI features, but the ones that attach AI to something competitors cannot copy in one release.