Stripe and Uber deploy AI agents to write company code
- Stripe Minions generate 30 percent of the company's merged code through one-shot prompts.
- Uber pairs engineers with business experts in two-week pods to automate manual operations.
- Accelerated agent deployments leave product design teams struggling to keep pace with engineering.
Major tech platforms are moving past AI text prompts to hand entire codebases and operational loops over to autonomous software agents.
At Stripe, internal coding agents are already writing nearly a third of the platform's code. On The a16z Show on August 17, 2026, Stripe's Will Gaybrick revealed that an internal tool called Stripe Minions generated 7,000 pull requests in a single week, accounting for 30 percent of all merged code changes. Instead of using these efficiency gains to trim engineering headcount, Stripe flattened management structures and directed developers toward shipping long-deferred features.
That agentic leverage enabled two developers to build Kai, Stripe's internal knowledge assistant, in just six months - a tool now used weekly by 83 percent of the company. Using agentic templates, Stripe also built its global tax filing infrastructure in one-third of the time it took to create its earlier, less complex US-only filing product.
"Stripe chose to build more, not hire less."
- Will Gaybrick, The a16z Show
The shift extends far beyond pure software engineering into core corporate operations.
On August 16, 2026, on The AI Daily Brief, host Nathaniel Whittemore detailed how Uber CTO Praveen Napali bypassed employee readiness barriers by launching two-week agentic pods. Uber paired 30 AI-proficient engineers directly with domain experts across legal, finance, and marketing. Over ten days, these pods shadowed workers, identified operational bottlenecks, and built custom autonomous tools.
The operational gains were immediate. Uber's capital allocation planning dropped from 15 hours to 30 minutes, financial pacing reports fell from two days to 10 minutes, and marketing quality assurance shrunk from two weeks to 50 minutes. Almost all Uber engineers now rely on local or cloud AI agents to author the vast majority of company code pull requests.
This sudden rush of automated software execution has created severe organizational friction elsewhere. Speaking on Lenny's Podcast on August 16, 2026, OpenAI Head of Product Design Ian Silber explained that while coding agents boosted engineering productivity up to 100x, design processes remain slow and non-binary. Testing hundreds of concepts and conducting user research cannot be compressed into an automated script, leaving design teams anxious as engineering deployments outpace them.
"This is the best time in history to be a designer."
- Ian Silber, Lenny's Podcast
To bridge that gap, OpenAI is pioneering adaptive interfaces that replace static chat textboxes with context-aware, interactive surfaces. The explosion in engineering speed is also flipping startup hiring dynamics. Early-stage founders who once assigned one designer to 15 engineers are now experimenting with two designers for every senior engineer, making human empathy and user experience the primary competitive moat.
As autonomous agents begin executing both software delivery and B2B commercial transactions, corporate leverage is decoupling from headcount. The burden is shifting from raw technical execution to human direction, strategy, and judgment.
Source Intelligence
- Deep dive into what was said in the episodes
Stripe’s AI Strategy: Build More, Not Less • Aug 17
- Stripe utilizes "Stripe Minions," an internal agentic tool that generates code via one-shot prompts. Will Gaborick reports that these minions generated 7,000 pull requests in a single week, accounting for 30 percent of Stripe's total code changes.
- To support hyper-productive engineers, Stripe is shifting toward flatter, smaller team structures. Will Gaborick highlights a single engineer orchestrating 16 agentic tools to build Stripe Projects, bypassing management layers to ship software faster.
- Two developers built Stripe's internal knowledge assistant, Kai, in six months. Will Gaborick notes the tool reached 83 percent weekly active usage and boosted sales representative productivity by 20 percent without reducing corporate headcount.
- Stripe uses agentic engineering templates to dramatically accelerate product development. Will Gaborick reveals that Stripe built its global tax filing product in one-third of the time it took to build the less complex US-only filing version.
- Traditional checkout pages will eventually disappear as AI agents take over internet purchasing. To enable this transition, Stripe launched the Link Agent Wallet CLI, allowing automated agents to leverage Stripe's 400 million Link user credentials.
- B2B transactions represent the most immediate opportunity for agentic commerce. Will Gaborick explains that Stripe Projects allows AI agents to provision and pay for cloud infrastructure services like Vercel directly, without relying on human interface navigation.
Also discussed on this episode: (9)
Safety (1)
- Will Gaborick describes Stripe as a multi-product platform spanning 25 to 30 core products. To block trial abuse for AI companies like 11 Labs, Stripe built a reasoning pipeline that identifies and stops automated fraud.
Payments (1)
- Digital goods companies use Stripe Managed Payments to scale globally without registering local business entities. Will Gaborick notes that Stripe acts as the merchant of record to handle tax compliance across more than 100 countries.
Startups (1)
- Will Gaborick states Stripe focuses on winning startups first because they demand higher software standards than enterprises. Startups find Stripe's reporting lacking compared to incumbents, pushing the team to continuously elevate its product quality.
Markets (1)
- Stripe experienced a 50 percent year-over-year increase in first-half signups. David George highlights that newer cohorts are growing rapidly, with the median 2026 cohort generating 50 percent more revenue than the previous year's cohort.
Enterprise (1)
- The demand for vertical SaaS software is expanding rapidly. Will Gaborick reports that Stripe's new SaaS platform cohort grew by 103 percent year-over-year, contradicting earlier market narratives that specialized SaaS platforms would struggle.
Stablecoins (1)
- Stablecoins offer superior global transaction efficiency by bypassing fragmented national rails. Will Gaborick notes that while Stripe's fiat currency network supports 60 countries, its native stablecoin integration expands Stripe's reach to 150 countries.
Protocol (1)
- Stripe is developing Tempo, a payment-specific blockchain built in collaboration with partners like DoorDash. The protocol prioritizes transaction privacy, high throughput during market surges, and stable transaction fees by avoiding floating gas pricing models.
Coding (2)
- As platforms like Cursor and Replit process massive token volumes, the line between software tokens and fiat money is blurring. Will Gaborick asserts Stripe must secure and manage token transactions with the same compliance rigor as traditional fiat.
- To maintain high software quality, Stripe simulates live customer environments for its engineering managers. Will Gaborick explains that generating mock data with disputes and refunds allows developers to experience real user friction and systematically address design flaws.
OpenAI’s Head of Design: This is the best time in history to be a designer | Ian Silber • Aug 16
- Ian Silber observes that while coding agents have boosted software engineering productivity by 10x to 100x, design teams have not seen matching gains. The design process remains highly iterative, messy, and dependent on user feedback.
- Lenny Rachitsky references Jenny Wen's observation that rapid development cycles have squeezed out traditional step-by-step design processes. Ian Silber notes that OpenAI embraces this shift by building in public and sometimes shipping features within four hours.
- Ian Silber plans to move ChatGPT beyond a simple text terminal by introducing context-specific, interactive elements like writing blocks. The long-term vision is a proactive, voice-enabled super app that adapts to a user's life context.
Also discussed on this episode: (7)
Labor (1)
- Lenny Rachitsky's sentiment survey of the tech workforce reveals that product designers and user researchers are currently the most unhappy cohort. They report the highest levels of anxiety, fatigue, and career pessimism.
Startups (2)
- Ian Silber argues this is the best historical era for designers because AI accelerates prototyping and empowers smaller teams. Startups are shifting from traditional ratios of 15 engineers per designer to hiring multiple designers per single engineer.
- During his time at Groupon, Ian Silber learned the value of injecting distinct brand character into products. Groupon succeeded early because its content was written by actual comedians, proving that a unique brand voice drives user engagement.
Models (2)
- Ian Silber claims AI is already an incredible product designer, but humans remain essential for inventing entirely new interaction paradigms. Great historical design shifts, like iPhone's multi-touch or Snapchat's camera-first interface, lacked prior training data.
- Managing ChatGPT's user base of over one billion active users requires balancing diverse needs, from casual domestic queries to highly technical farm automation. Ian Silber uses early desktop app experiments to test capabilities before simplifying them for the masses.
Reasoning (1)
- Ian Silber emphasizes systems thinking in product design, citing Notion's composable building blocks as a model. Creating unified primitives allows underlying AI models to easily reason about and navigate the entire user experience.
Social Media (1)
- Reflecting on his eight years at Instagram, Ian Silber notes how the failure of IGTV paved the way for Reels. Success in fast-moving industries requires abandoning rigid assumptions and fixing forward after public flops.
Nathaniel Whittemore
The New Problems AI Is Creating (And How People Are Solving Them) • Aug 16
- Engineering workflows at Uber are heavily integrated with AI. CTO Praveen Napali states that nearly all company engineers use AI tools, with local or cloud agents generating the vast majority of code pull requests.
- Uber accelerated operational tasks by pairing AI-proficient engineers with domain experts in two-week agentic pods. This collaborative approach redesigned entire workflows, radically slashing times for capital allocation, financial pacing reports, and quality assurance testing.
Also discussed on this episode: (6)
Agents (1)
- While organizations rush to implement AI agents, employee readiness lags significantly. Section's report reveals that despite widespread corporate interest, very few workers actually use these tools or even understand what they are due to a widespread lack of training.
Enterprise (2)
- AI adoption is intensifying work density rather than liberating free time. ActiveTrack research shows that early adopters experienced doubled messaging volumes, increased business software usage, and a decline in focused, uninterrupted work time.
- Low standards plague corporate AI adoption as workers optimize for speed over quality. A GoTo survey found that nearly half of employees admit to submitting AI-generated work despite suspecting it contained errors or was of low quality.
Coding (1)
- Haas School of Business researchers found that AI enables employees to tackle specialized tasks like coding that they previously outsourced. This technical empowerment leads to fragmented work bursts during evenings and weekends, causing severe multitasking and mental fatigue.
Brain (1)
- David Brooks warns that relying on AI to bypass mental effort risks degrading critical thinking skills. MIT Media Lab and Possibility Sciences research both show marked declines in brain connectivity and gamma wave activity when individuals utilize AI for tasks.
Psychology (1)
- Nathaniel Whittemore argues that the true value of AI lies in attempting tasks previously beyond one's capability, rather than just automating existing routines. Navigating technical friction and building complex tools expands personal ambition and cognitive elasticity far more than optimization.