The Benchmark You've Been Missing
Most internal AI conversations stall on the same question: how long will this actually take? Vendors say weeks. Engineers say quarters. Nobody has a clean reference point.
Roman Ugarte's case study on building Grok Bot in one month is the closest thing to a calibrated answer the industry has produced. It's a real PM documenting a real build — constraints, tradeoffs, decisions included — not a retrospective polish job. The timeline is 30 days from zero to production. That's the benchmark.
For operators at non-tech companies deciding whether to build an internal AI agent this quarter, that number changes the conversation.
What the Timeline Actually Contains
One month sounds fast until you break it down. What Ugarte's account reveals is that speed came from scope discipline, not heroics. The team made deliberate choices about what the bot would not do. They scoped to a single, high-value workflow. They didn't try to build a general assistant.
This is the part most internal AI projects get wrong. The instinct is to solve for flexibility — build something that could handle anything. That instinct is expensive. A focused agent with a defined job ships in weeks. A flexible platform ships in never.
The other factor is tooling. The current AI development stack has compressed what used to require months of custom infrastructure into configuration work. That's not a reason to underestimate complexity — it's a reason to reassess your timeline assumptions before the next planning cycle.
The Build vs. Buy Question Has a New Default
For most non-tech companies, the default has been to buy: procure a vendor solution, configure it, move on. That default made sense when building meant a six-month engineering project. It's less obvious when a focused internal build takes 30 days and produces something shaped exactly to your workflow.
The honest answer is still: it depends. Vendor solutions make sense when the use case is generic and the vendor has already solved for compliance, integrations, and support. Building makes sense when the workflow is specific, the data is sensitive, or the competitive value is in the customization.
What Ugarte's case study does is move the breakeven point. The build option is now credibly on the table for a wider range of use cases than it was 18 months ago.
Actionable takeaway: Before your next AI vendor evaluation, run a 30-day build sprint as a parallel track. Scope it to one workflow, assign one owner, and use the output to pressure-test the vendor's timeline and cost assumptions. Even if you end up buying, the sprint will make you a sharper buyer.