Scuttlebutt

Scuttlebutt

Devin's own Android app was built by one AI agent — Cognition's co-founder and CEO

4 takeaways: why "not a single human" worked on it, the 10-to-30x cost gap agents now route around, and why he calls point-in-time security scans obsolete

Scuttlebutt's avatar
Scuttlebutt
Nov 19, 2026
∙ Paid

TBPN · Nov 18, 2026

With Scott Wu, Co-founder and CEO of Cognition

Devin's own Android app was built by one AI agent — Cognition's co-founder and CEO

Photo via TechCrunch

Scott Wu, co-founder and CEO of Cognition, told TBPN that the company's own Android app was built entirely by a single, standing AI agent.

Where most AI coding tools still wait for a person to direct them, Scott describes Cognition's newly launched "Devin Bots" as agents that own outcomes and projects on their own.

"There was not a single human who worked on that." — Scott Wu

Scott co-founded Cognition and has built the company around proprietary data from how people actually use its Devin coding agent.

The full segment is covered here so you can skip it.

Here are the 4 takeaways that matter.


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • Cognition's own Android app was built entirely by one persistent "Devin Bot," with no human involved

  • Routing between models that differ in cost by roughly 10 to 30 times is now central to making agents economical

  • Cognition hired former Facebook CSO Alex Stamos and argues point-in-time code scans are obsolete against AI-driven attacks

  • Scott expects a shift from today's "golden era" of tool use toward self-driving, personalized software

1. A Bot That Owns the Outcome

Scott opened by describing what makes Cognition's new "Devin Bots" different from earlier coding assistants: they own a project end to end rather than completing one task at a time.

Scott Wu says Devin's own Android app had no human contributor

"There was not a single human who worked on that." — Scott Wu

He pointed to a second example as proof the model generalizes beyond one app: an "always-on Devin" that continuously works to optimize inference kernels on its own, without a person assigning each change.

2. A 10-to-30x Cost Gap

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.