Michael Quoc
@michaelquoc
Followers
6K
Following
22K
Media
381
Statuses
1K
Founder/CEO @demandio | $20M+ ARR, 20 people, zero VC | Building Truth Engines, not Relevance Engines | Codifying AIOS & Context Engineering in public
Santa Monica, CA
Joined January 2008
We run $20M+ ARR on 20 people by treating context as infrastructure, not prompts. After 18 months building AIOS (our internal AI context engine) in production, we've codified 7 axioms that govern our entire context architecture. These aren't theories. They're constraints we
1
1
2
@demandio For the full technical deep-dive on Context Engineering (4,500 words, production architecture, failure modes, implementation patterns): https://t.co/DnDcw8FAoB This is the canonical resource. The thread is the executive summary.
thefirstprinciple.io
Why the "Prompt" is a Mirage and How to Build a True AI-Native Operating System
0
0
0
These 7 axioms govern how we architect AIOS at @demandio. They're constraints from production, not academic theory. But we're not building in a vacuum. Which axiom breaks first at 10x scale? Which constraint are we wrong about? Debate in replies. Show your work.
1
0
0
Axiom 7: Culture Owns the Context Portfolio. The Hardest Part: Technical stacks fail without cultural accountability. Our WIP Model: "Data as Product." Human owners are accountable for their context domain. This is our shared brain (LCP). Still prototyping. Still learning.
1
0
0
Axiom 6: RAG is CI/CD for Truth. Production Lesson: Context systems drift. Fast. Our Eval Protocol: Continuous A/B testing with RAGAS (faithfulness) and OWASP (red team). We measure retrieval precision and hallucination rates daily. Deploy-and-forget RAG is deploy-and-fail.
1
0
0
Axiom 5: The Context Pipeline is Your Attack Surface. Most "hallucinations" are temporal bugs (stale data) or injection attacks. Our AIOS Design: Zero-trust RAG firewalls + bitemporal modeling. Every context fragment has provenance and validity timestamps. Without governance,
1
0
0
Axiom 4: Compression is Non-Negotiable. Data: 50-95% of KV cache is redundant. Uncompressed context murders TTFT. Our R&D Focus: Positional-aware compression. We've hit 5-50% footprint with zero accuracy loss and 2x throughput. Prefill latency is the primary UX bottleneck.
1
0
0
Axiom 3: You Need 6 Primitives, Not One. Most builders treat context as "just RAG." That's why most RAG systems hallucinate. Our AIOS Stack: Bitemporal identity → Entity graph → Semantic layer → Vector index → Policy graph → Tool graph. Missing any primitive = structural
1
0
0
Axiom 2: Context is RAM, Not Storage. @karpathy had this insight: LLM = CPU, context window = volatile memory. Our Production Reality: We measure context density (ρ) in outcome-lift-per-token. Dense context = 2-5x better results with same model. Sparse context = expensive
1
0
0
Axiom 1: Prompting is Vibe Coding. The Problem: One-shot prompts have zero memory. Every interaction starts from amnesia. Our AIOS Constraint: We architected persistent context pools that survive across sessions. Without state, you're just rebuilding context from scratch every
1
0
0
Use cases like this are inspiring and provide a glimpse into how the world will experience crypto in the future. We won’t even call it crypto at that point. We’ll just have access to new ways to connect, communicate, and transact on our phones, secured by a blockchain. I think
this is why i love crypto so much. fred again is playing his first stadium show, the biggest of his career after his meteoric rise. before the show, he personally tracked down the 150 people that attended in 2021 before he blew up and invited them to a secret listening party.
1
1
16
Exciting news! Say hello to the revamped SimplyCodes. Brighter, bolder, and better than ever before to help you experience even more discount codes. Check it out at https://t.co/U0GhllBhFA.
#discounts #SimplyCodes #savemore
1
4
15
Significant progress towards AGI with Google Gemini coming soon.
I’m very excited to share our work on Gemini today! Gemini is a family of multimodal models that demonstrate really strong capabilities across the image, audio, video, and text domains. Our most-capable model, Gemini Ultra, advances the state of the art in 30 of 32 benchmarks,
1
0
4
👀👀👀
“ Ahead of OpenAI CEO Sam Altman's four days in exile, several staff researchers sent the board of directors a letter warning of a powerful artificial intelligence discovery that they said could threaten humanity, two people familiar with the matter told Reuters. “
0
0
2
Something big happened in AGI behind the scenes at OpenAI that probably makes all this external drama look like child's play. @karpathy also tweeted this amid all the OpenAI employee 💙 tweets
@DrKnowItAll16 I am very worried. Ilya has a good moral compass and does not seek power. He would not take such drastic action unless he felt it was absolutely necessary.
0
0
2
The power games in AGI are high stakes at an unprecedented level 👀
It seems like Altman was leveraging his OpenAI streetcred to build new personal AI companies (a compute one and a device one), which: 1. would provide him with potentially enormous amounts of bargaining power with OpenAI, especially the compute one. 2. would make him much more
0
0
0
This feels to me like the most accurate assessment of really happened behind the scenes in @sama and @gdb exit from @OpenAI
OK everyone's asking me for my take on the OpenAI stuff, so here it is. I have a strong feeling about what's going on, but no internal info so this is just me talking. The first point to make is that the Dev Day was (IMO) an absolute embarrassment.
0
0
0
Concerning that the CEO of the company that’s bringing AGI to the world is being accused of ethical issues by his own board
openai.com
0
0
1
Well said 🚀🚀
more has happened in the last year of AI than the prior 10 we are unmistakably in the fiery takeoff of the most important technology of the rest of our lives everybody—governments, citizens, technologists—is awaiting w/bated breath (some helplessly) the next version of humanity
0
0
1