a16z opens a $42M tuition-free academy as a college alternative
Andreessen Horowitz spent Tuesday arguing that the pipeline into AI work needs rebuilding from the school up — and put $35 million behind the claim.
a16z launched the Horowitz Andreessen Academy, an unaccredited San Francisco program that trades degrees and homework for a year of building, and started taking applications the same day. The firm contributed $35 million of the $42 million raised; the other $7 million came from founding partners that read like a frontier-industry roster — Anduril, Anthropic, Coinbase, Google, Meta, Nvidia, OpenAI, Palantir, Replit and Stripe. The first cohort, the Founding Class Fellowship, is tuition-free, takes about 50 students aged 16 to 22, and starts in fall 2027. Udemy co-founder Gagan Biyani runs it, with Marc Andreessen and Erik Torenberg joining him on the board.
The numbers attached to the students are the ones worth reading twice. Each one gets access to more than $50,000 worth of AI compute from the partner companies, along with their networks and advice on the curriculum — a per-head compute budget most university research groups would envy. More than 40 companies, including Applied Intuition, Databricks and Shopify, have signed on as hiring partners to scout interns and full-time hires. Instructors will not need teaching credentials or a PhD; the tracks on offer include "AI Research & Models," "Aerospace, Defense & Industrial Systems" and, less predictably, "Writing & Culture."
The straightforward read is that a16z is building a recruiting pipeline with a campus attached. The academy is set up as an independent business rather than a school, which keeps it outside accreditation rules while the partners get early access to a hand-picked cohort — and the compute stipend is the part of the offer a university cannot match on its own. Critics of AI-era schooling have warned that industry-led instruction puts credentials at risk; the academy's answer is that the credential is the job. "The training that worked for the Industrial Revolution isn't going to map perfectly onto the AI revolution, so somebody has to pioneer how you train a person for this new world," a16z co-founder Ben Horowitz said. That is a real gap in the market and a real experiment — but the offer is also a bet that AI companies will keep paying a premium for people who already know how to build.
TypeSafe's decision models now plug into the LLM command-line tool. Simon Willison released llm-typesafe 0.1a0, a plugin that puts TypeSafe's Jev model behind the interface developers already use for other models — and returns typed answers instead of prose: a yes/no probability, one of several named categories with per-category confidence, or a score on a rubric you define. Structured JSON input and reusable saved templates come with it. Jev's whole pitch is that software cannot consume sentences, and this plugin is the piece that lets a pipeline call it without anyone writing custom glue. We covered TypeSafe's exit from stealth — TypeSafe exits stealth with $40M for models that don't chat.
What to watch: whether the academy pursues accreditation, and whether the partner compute promises ever show up as published numbers per student.
Would you trade a degree for a year of building on someone else's compute? Tell us in the comments.
Sources: llm-typesafe (GitHub)
