TypeSafe AI CEO Diogo Almeida says established business-software companies have an edge in AI automation because they understand customers’ workflows and can fund the engineering to automate them. Distribution, in his telling, isn’t the magic ingredient. Source
Speaking on 20VC in an episode released November 16, 2026, Almeida also stopped short of declaring victory for software-as-a-service companies, or SaaS. He sees an advantage for incumbents, not a trophy reserved for whoever already has the most customers. Source Source
“It takes real engineering to automate stuff,” he said. His hypothetical example was improving TurboTax’s interface: teams familiar with the product know where changes could make the work smoother. He contrasted that proximity with OpenAI and Anthropic, which he criticized as “still living in their ivory tower.” Source Source
Almeida has a commercial stake in this argument. TypeSafe makes Jev, which 20VC’s episode description presents as a model for automated software decisions that reports its confidence in each call. Source
Being online isn’t the same as being right
By reliability, Almeida means an AI model consistently making sound decisions, not simply a service staying online. He called this “reliability of intelligence” and identified it as the central obstacle to software that can work unattended. Source
In Almeida’s account, unreliable models discourage companies from putting AI deep into consequential application logic. Instead, they add an assistant and leave users responsible for checking its output. The automation pitch comes with homework. Source
The model isn’t the whole business
Almeida also argued that proprietary intelligence can come from software built around models, rather than every company training its own. In his view, differentiation can happen “at the code level”: how a business turns a model’s capabilities into working automation. Source
Comments