HiBob turned 2,500 GPT experiments into a 90% adoption rate

Most companies hand out ChatGPT Enterprise logins and pray for the best. HiBob decided that was a recipe for a very expensive chatbot. The HR tech firm, which serves over 5,000 customers, didn’t stop at granting access. They built a five-step assembly line to turn raw ideas into what they call ‘digital companions,’ and the numbers are frankly a little jarring: 2,500 experimental GPTs were created, 200 made it into production, and over 90% of the company now actively uses the platform.

That’s not a typo. That’s a cultural shift, engineered from the top down by an internal AI Mind team. The framework requires employees to tie every GPT pitch to a concrete pain point before an engineer touches it. Successful agents get a named owner, documentation, and a feedback loop. If they prove their worth, they’re added to a searchable internal directory for other teams to steal. It’s DevOps for office work, transforming disposable chatbots into versioned, maintained internal products.

The scope is refreshingly practical. A ‘Meeting Prep GPT’ pulls CRM data and transcripts to build briefings, saving sales and CS teams hours of grunt work each week. An ‘Upsell GPT’ scans usage patterns to flag revenue expansion opportunities. These aren’t science fair projects. They’re tools that chip away at the relentless, tedious tasks that make people dread their calendars. HiBob’s approach treats an agent like a new hire who needs a defined role and a manager, not just a prompt.

Crucially, this internal sandbox directly feeds the product roadmap. The company prototypes customer-facing features inside ChatGPT Enterprise, stress-tests them, and then implements them via OpenAI’s API on the Bob platform. It’s a closed loop where dogfooding isn’t just a buzzword; it’s the R&D engine. The bet is that letting employees own the process doesn’t just boost adoption—it makes the AI smarter faster. But the real test is whether this pace is sustainable once the novelty fades and every team is fighting for compute and engineering time.