Back
You don't need a full ML team to ship an AI product anymore. What you need is a clear-eyed view of where AI creates durable value for your users, and a team that knows how to integrate foundation models into real workflows.
What's actually changed
Foundation models from Anthropic, OpenAI and Google now handle most of the modelling work that used to require dedicated researchers. What remains — evaluation, guardrails, integration, data plumbing — is product and engineering work. That shift is why small teams are now shipping AI products that would have required a research lab five years ago.
The new skills that actually matter
Prompt design, eval pipelines, structured output, retrieval architectures, and cost-aware orchestration. None of these require a PhD — they require product engineers who take AI seriously and aren't afraid to read papers when the situation calls for it.
The teams moving fastest right now are the ones who treat AI as a tool to sharpen existing workflows, not a novelty to bolt onto the product.
What to build first
Start with the unglamorous places where your users waste time — approvals, data entry, triage, classification. Measure the before and after. Ship small, learn fast, then go wider. The companies doing this best are the ones who treat AI as infrastructure, not as a headline feature.
