👈👉 Click arrows or swipe to navigate
1 / 21
"We are not stuff that abides, but patterns that perpetuate themselves."
- Norbert Wiener, Father of Cybernetics
Ast. AI Factory Foundry
by Promptwerk Incorporated
Palo Alto, California, USA
June 2025
Stanford EE + Cal CS Team
  • 🧑‍🔬 Engineers, researchers, builders
  • 🙌 Campus-born innovation
Boomer + Gen Z Dream Team Median: 22 years exp
Serious skills meet fresh energy
(And, yeah, we know it's wild)
Recursive AI: AI Bootstraps AI
  • Bootstrapped 100% of coding - with AI - since 2024
  • Way before the "vibe" hype hit the fan
$500K SAFE raised (Apr 2024)
Under $100K burned - Lean by design
Solving the Russian Roulette to Six Nine Problem
Why risk failures when you can build in reliability?
The Ast. Recursive AI Foundry
  • Builds safe, reliable, secure AI factories
  • Factories build other factories - it's recursion all the way up
Factory Isolation = Safety First 🥜
  • Factories run isolated from both foundry & others
  • (Peanut allergy: no cross-contamination)
We Respect Copywrites & Rights
  • Hundreds of models, 1000s of workflow steps
  • Licensed content, copywrite-protected, full audit trail
  • Easy rights-holder payouts (like YouTube Content ID)
Workflow Engine: Parallel & Recursive
  • Parallel & synchronous steps
    Recursive loops, no limits
Danger: Model Swaps Break Stuff!
Models improve - but swapping fast triggers new errors. Never skip thorough testing!
(Every model makes different mistakes!)
TL;DR Summary
  • LLMs build factories
  • Factories come with safety rails
  • Humans jump in only for edge/high-stake cases
  • System improves autonomously
  • Extreme scale, speed, trust + reliability
Six Nines Reliability in an AI-First World
Accountable, auditable, safe automation
- at massive scale
How Factories Are Built
  1. Expert Team x LLMs
  2. Experts architect & approve each pipeline
  3. LLMs draft, optimize, generate code & tests
  4. Automated QA & compliance built in
User? Just input your goal in natural language
Secure NLP Inputs Only
  • User gives easy natural language input
  • Via trusted, authenticated channel
  • No prompts to engineer, no config needed
Expert + AI
Continuous Feedback
  • Edge cases escalate to humans
  • LLMs suggest fixes, experts approve/implement
  • Factory gets smarter & safer with every run
Why Not Just Use Chat?
LabelPure LLM ChatDIY User FactoriesExpert-Led, LLM-Powered Factories
ExpertiseNone (user)User (low/medium)Expert (high) + LLM (speed)
QualityVariableVariable, brittleHigh, consistent
User EffortHighMediumVery Low
TrustLowMediumHigh
ComplianceUnreliableRiskyBaked in
AdaptivityHigh (risky)MediumHigh, controlled
Horizontal scroll on mobile
User Flow 🚀
  1. User logs in (secure)
  2. Gives simple, natural request
  3. Factory interprets, validates workflow
  4. Automated QA/compliance checks
  5. Edge cases go to experts
  6. Reliable output, every time
  7. System learns from feedback
The Takeaway
  • LLMs + Experts = Unmatched reliability
  • Users: Simple, safe, automated
    No prompt engineering, no hacks
  • Factories you can audit, trust, scale
Thanks, craigw@ast.engineer! 🌟
"Happiness is not an ideal of reason, but of imagination." - Immanuel Kant
made by ast