Aidan Gomez and Cohere: Skipping the Chatbot Race
One of the authors of the transformer paper runs a company that deliberately stayed out of the consumer race. He is selling to the procurement department instead.
I have never met Aidan Gomez, and I am told he was a graduate student when he helped write the paper that introduced the transformer architecture, which means a large fraction of everything now described as artificial intelligence traces back to work he did before he had a mortgage. I have known a good many men present at the creation of something, and few kept quiet about it.
What makes his company interesting — and I say interesting the way a man says it about a horse that refuses to run the obvious race — is not that pedigree but the market it chose. Cohere sells to businesses, not consumers, and has largely avoided the assistant arms race that consumes its better-known competitors.
The unglamorous market
Selling retrieval, generation and classification tools to enterprises is a slower business than launching a chatbot. Deals take months, procurement is bureaucratic, and the customer’s success metric is often cost per document rather than any kind of benchmark score.
It is also a market where being slightly behind on raw capability matters far less than deployment flexibility: running inside a customer’s own cloud, meeting data-residency requirements, and being auditable. Nobody ever won a conference announcing that the auditors were satisfied.
Why the strategy makes sense
- Enterprise contracts are stickier and less exposed to consumer fashion.
- Deployment flexibility is a genuine differentiator against platform-owned models.
- Regulated industries need a vendor who will sign a data processing agreement.
The pressure that comes with it
The risk of the enterprise-only strategy is that the platform vendors can bundle. If a company is already paying for a cloud and a productivity suite, an adequate model included in the price is often good enough, and no committee ever lost its job for choosing the thing already on the invoice. Cohere’s defence is depth: better retrieval, better deployment, and better support for the workflows enterprises actually run.
Gomez has been consistent that the company is not trying to win a general intelligence race. In a field full of maximalist claims, that restraint is itself a positioning statement — and the sort of modesty that costs nothing and pays regularly.
What enterprise buyers actually ask for
The questions that determine whether a deal closes are rarely about benchmark scores. They concern where the data is processed, who can access it, what happens during an audit, the escalation path when something fails at two in the morning, and whether the vendor will still exist in five years.
Answering those questions is unglamorous work that requires legal, security and support organisations rather than research breakthroughs. It is also why enterprise contracts are sticky: replacing a vendor means repeating the entire evaluation, and no lawyer has ever been talked out of a form.
The commoditisation risk
The structural danger is that the capability layer becomes a commodity and the value migrates to whoever already owns the customer relationship. A company selling models to enterprises is ultimately competing with the cloud provider that already runs the enterprise’s infrastructure — and that provider holds the keys and signs the renewal.
The defence is specialisation and depth in the workflows that matter most. It is a narrower position than being a frontier lab, and a more durable one if the capability race eventually stops producing decisive advantages. I have watched a great many rivers change course, and never yet seen one regret a deep channel.
Image credit and licence details for every photograph on this site are listed on the credits page. This article is editorial content; it carries no sponsored material.