John Collison: agentic commerce, "math wars" and Stripe's rebuild for AI
Key points
Key takeaways from a Sourcery with Molly O'Shea interview with John Collison, co-founder of Stripe, recorded on stage at Browserbase Navigate (October 2026):
Payments at scale. Stripe processed $1.9 trillion in 2025, up 34% year on year, and Collison argues that after years of stability the payments industry is finally facing its biggest reset since the company's founding, in its 17th year.
Agents replace web forms. Computer use is "the ultimate backwards compatibility layer": once personal agents hold stateful context, such as shipping address and payment details, and can operate a computer, users can ask an agent to renew a driver's licence or run an errand rather than filling out forms — no need to wait years for the DMV to ship an MCP.
Commerce rewiring. Agentic commerce is a "complete rewiring": AI discovery becomes the new point of aggregation, so well-reviewed niche brands gain reach and existing aggregators face questions — analogous to the Facebook and Instagram ad-targeting era that spawned native brands such as Wish.com, Teespring, Temu and Shein.
Cyber risk rises. The new models arm defenders and attackers alike, and Collison expects more breaches in the next five years than in the past five; well-resourced incumbents running 50-year-old, acquisition-grown tech stacks are most exposed, while startups' smaller, contained, homogeneous stacks let them adopt the latest capabilities faster.
Singularity epoch. Stripe half-jokingly designates 1 January 2026 as "day one of the singularity"; with labs racing on recursive self-improvement — he calls the current model-lab sprint "math wars" — Stripe sees the highest rate of new firm creation in recent memory across its platform.
Rebuilding for AI. Almost nobody integrates the Stripe API by hand any more — coding agents do — and "developer ergonomics are different from AI ergonomics", so Stripe is changing what it is good at internally rather than polishing developer-facing details.
Acquisition strategy. Bridge and Privy injected crypto-native culture, Metronome added usage-based billing for AI-era pricing, and OpenRouter brings multi-model routing on the belief that businesses will run several models at once; Stripe buys a "new muscle" in a fast-moving market and builds where it is already core.
AI inside Stripe. Internal AI applications with strict access-control guardrails now sit on Stripe's own data: sales teams self-serve personalised pitches by building their own apps, and the legal team runs an agent that ingests new regulation, such as 600-page laws in India and Finland.
Read more: Sourcery with Molly O'Shea on YouTube
