TL;DR
- Brex scrapped its traditional software procurement strategy due to slow internal timelines
- The company created a new framework for faster AI tool vetting and legal clearance
- Employees now receive a personal monthly software budget to license AI tools
- Brex evaluates tool success using a “superhuman product-market-fit test”
- CTO James Reggio encourages enterprises to “embrace the messiness” of AI adoption
Traditional Procurement Fails to Match AI Speed
Brex, the corporate credit card and expense management startup, found itself facing a dilemma familiar to many large enterprises: how to keep up with AI innovation when procurement processes can’t move fast enough.
Speaking at the HumanX AI conference in March, Brex CTO James Reggio explained that the company’s original procurement system involved months-long cycles of vetting, piloting, and approval. This structure, designed for more static SaaS products, simply couldn’t keep pace with the dynamic AI tools entering the market after the launch of ChatGPT in late 2022.
“In the first year following ChatGPT, when all these new tools were coming on the scene, the process itself of procuring would actually run so long that the teams that were asking to procure a tool lost interest by the time we got through all of the necessary internal controls,” said Reggio.
Building a New AI Tool Framework
To address the lag, Brex completely restructured how it evaluates and procures AI tools. The company began by designing a new framework for data processing agreements and legal validations specifically for AI-powered software. This internal structure allowed teams to pilot new tools faster, without compromising compliance or security standards.
Once new AI tools passed the preliminary legal and technical review, they were handed off to employees for field testing — but with an important twist: Brex now involves engineers directly in procurement decisions.
Brex’s AI Procurement Shift
| Category | Detail | Source |
| CTO Insight | Procurement cycles too slow post-ChatGPT | TechCrunch |
| Strategy Shift | Custom legal framework for AI tools | TechCrunch |
| Employee Involvement | Monthly software budgets for engineers | TechCrunch |
| Evaluation Metric | “Superhuman product-market-fit test” | TechCrunch |
| Tool Volume | Over 1,000 AI tools piloted | TechCrunch |
“Superhuman Product-Market Fit” as the Evaluation Model
Rather than relying on executive decisions or long-term procurement projections, Brex now bases its software investments on a “superhuman product-market-fit test.” This means tools are retained or expanded based on actual value seen by frontline users, not executive preference or marketing hype.
“We go deep with the folks who are getting the most value out of the tool to figure out whether it is actually unique enough to retain,” said Reggio.
This shift has also enabled faster iterations. Brex has tested more than 1,000 AI tools in the past two years and has already dropped five to ten larger-scale deployments that didn’t deliver on their early promise.
Monthly Software Budgets Empower Engineers
One of Brex’s most radical internal changes is giving engineers a $50 monthly software budget to license whichever tools they find useful — provided they come from a pre-approved vendor list.
By delegating purchasing power to users, the company avoids bottlenecks and gains deeper insights into which tools actually improve workflow.
“It’s actually really interesting — we haven’t seen a convergence,” Reggio said. “That has validated the decision to make it easy to try a bunch of different tools. We haven’t seen everybody just rush in and say, ‘I want Cursor.’”
This decentralized approach also helps Brex identify when broader enterprise licensing deals are warranted, based on organic adoption patterns, not forecasts.
Licensing, Tool Sprawl, and Iteration
Rather than fear tool sprawl, Brex has embraced it as a natural phase of innovation. By monitoring tool adoption through its decentralized budget model, the company can scale up promising technologies while abandoning underperformers quickly.
Reggio views this agility as an asset: “Knowing that you’re not going to always make the right decision out of the gate is just paramount to making sure that you don’t get left behind.”
Embracing the Messiness
Reggio’s core advice to enterprise leaders is simple but contrarian: embrace the messiness.
AI innovation cycles are too fast — and too unpredictable — for rigid workflows and overengineered evaluations. Enterprises that attempt to vet every possible tool with six- to nine-month decision cycles will almost certainly find themselves obsolete by the time a decision is made.
“The one mistake we could make is to overthink this,” he said. “You don’t know what the world is going to look like nine months from now.”
Conclusion: Speed and Flexibility as Core AI Competencies
Brex’s overhaul of its AI procurement process shows that speed, adaptability, and employee trust may be more important than technical expertise when navigating the modern AI ecosystem. Involving engineers directly in tool testing and licensing empowers decision-making at the edge, where the real value of AI is measured.
Other enterprises may find themselves forced to follow suit — not out of preference, but out of necessity. In a world where tool relevancy shifts monthly, agility may be the only lasting competitive edge.
