What engineers spend on AI and what it buys - Tustin AI Daily Briefing, Oct 21

Tustin AI Daily Briefing · Oct 21, 2026 · 7:16

Listen in the Podli app 🎧

Follow your favourite podcasts, listen offline and in the car with CarPlay and Android Auto, and always pick up where you left off. Free to try.

Stories in today's episode — tap a headline to read the full piece:

What engineers actually spend on AI — and what it buys
DeepSeek open-sources DeepInfer, its production inference stack — and discloses the economics
Leaked Anthropic draft prospectus: $16.3B revenue in H1 2026, and a compute supply problem
Hackers used Claude and Grok to fully automate attacks on 400+ financial targets
Anthropic's own workforce data: junior technical hiring down from 27% to 9%, agents idle 80% of the time
Legora raises $550M at $8.8B as vertical AI and agent infrastructure keep clearing
SF Compute's Evan Conrad: why a compute market needs physical settlement
Reflection ships Beam-1.0, the strongest open American model under a permissive license
Someone has to own the AI data-center credit risk
Claude Mods turn the agent harness into a programmable "attention interface"
Every killed its personal-agent product and rebuilt around one shared company agent
The anatomy of an AI pivot: Intercom walked away from $60M in contracted ARR
Mollick: the METR long-horizon benchmark ran out of room, and answer costs collapsed 99%+
SpaceX and Tesla boards form special committees to review a merger
Per-token pricing still rules, so every AI startup gets benchmarked against the labs
Mollick charts 30 months of exponential AI progress — and Opus 5.5 answers an image prompt with a 110-second video
Khosla: one research paper could make models 10x faster — Brad Porter and John Giannandrea on moving AI from research to deployment
An AI concierge called a store on a user's behalf — with no transcript, no recording, and no disclosure
OpenAI, PerturbAI and Scripps run thousands of agents over biological data to generate research leads
How OpenAI built ChatGPT's ad stack — $1B ARR run rate in under 200 days