The Curious Engineer

All blogs / AI / ML

The AI Frontier

frontierai.substack.com

Lessons from building an AI product at RunLLM and the latest AI research at Cal. Click to read The AI Frontier, a Substack publication with thousands of subscribers.

AI / MLTech news / industry

’23’24’25’26
Posts
142 from 16 Aug 2023 to 3 Sep 2026
Posting
weekly (36 in the last year)
Kind
personal blog
Feed
https://frontierai.substack.com/feed

2026 30

  1. LLMs are becoming commodities
  2. Model inference, model products, and AI applications
  3. The SaaS Extinction Test
  4. Can you know if coding agents are worth the cost?
  5. Data is your only moat
  6. Open models don't need to be OpenAI
  7. AI changed distribution. Can you keep up?
  8. Budgeting for AI isn't hard
  9. AI is not a line item
  10. If you don’t know what good is, AI won’t tell you
  11. Enterprise AI's teenage years
  12. Product-market fit is a trap
  13. AI applications have two growth curves
  14. The Inference Economy: Token Use
  15. Agents can't choose between structure and flexibility
  16. Build agents for humans
  17. Agents can't check their own work
  18. AI agents shouldn't have a job title
  19. AI Companies Are Building for the Wrong Users
  20. Generic enterprise AI is indefensible
  21. The SaaS Extinction Test
  22. The zero-shot trap
  23. SAT Scores vs. On-the-Job Performance
  24. The case for wasteful agents
  25. Cursor's Unfair UX Advantage
  26. How to build a data moat
  27. What Moltbot tells us about the ideal UX for AI
  28. The specialization moat
  29. Data is your only moat
  30. The business of AI in 2026

2025 50

  1. Looking back on AI in 2025
  2. Maybe AI is a regular platform shift
  3. The rise of AI forward deployed engineers
  4. The 9-9-6 Trap
  5. The Inference Economy: Part II
  6. The Inference Economy
  7. Build opinionated agents
  8. Agents, knowledge debt, and asynchronous work
  9. Depth over breadth
  10. OpenAI is a consumer company
  11. You can't build a moat with AI
  12. A theory of the AI market
  13. Looking ahead in AI
  14. Lessons learned from a 100 blog posts on AI
  15. AI Artists vs. AI Engineers
  16. Model inference, model products, and AI applications
  17. Do you have too many AI products?
  18. Slack is the future of agentic work
  19. Does AI quality matter?
  20. Introducing RunLLM v2
  21. The new UX for AI is not what you think
  22. The AI Application Land Grab
  23. What happens when an AI is your customer?
  24. An AI Support Engineer for every open-source project
  25. Why hasn’t AI changed everything
  26. Will long context windows solve all your problems?
  27. A deep dive on search for AI applications
  28. LLMs on the Run
  29. What's an agent, anyway?
  30. Stop treating frontier models like products
  31. Design AI to fail
  32. Qwen 3 and doubling down on open-source AI
  33. AGI is already here
  34. Acquisitions, consolidation, and innovation in AI
  35. Your AI application needs a job
  36. Building AI applications you can trust
  37. Will OpenAI eat our lunch?
  38. AI is (still) all about data
  39. How to build an AI application that people trust
  40. What is an AI application?
  41. LLMs are too helpful
  42. The hard thing about building AI applications
  43. Introducing RunLLM: The first AI Support Engineer
  44. You can’t build a moat with AI (redux)
  45. So you want to buy your first AI product
  46. DeepSeek, o3, and AI applications
  47. AI is yet another platform shift
  48. One month of using Devin
  49. Chatbots are dead, long live chatbots!
  50. Predictions for AI in 2025

2024 45

  1. Looking back on AI in 2024
  2. Compound AI is AGI
  3. The end of scaling laws doesn't matter
  4. Your AI strategy is a waste of time
  5. AI lets you scalably do things that don't scale
  6. Can I talk to an AI, please?
  7. Compound AI, test-time compute, and wasting your users’ time
  8. Throw more AI at your problems
  9. A theory of the AI market
  10. Deep AI work
  11. Customers want more AI, not better AI
  12. Should AIs try to be human?
  13. AI companies should be AI-first
  14. AI workers will ignore your Slack messages too
  15. AI products should be built for human coworkers
  16. The rise of AI work
  17. The future of AI pricing
  18. Introducing the AI Frontier
  19. In defense of vibes-based evaluations
  20. There's no silver bullet in AI
  21. OpenAI is too cheap to beat (redux)
  22. Mid-2024 Predictions Review
  23. AI applications are more than an LLM
  24. The bull case for AI startups
  25. AI is breadth-first, not depth-first
  26. LLMs are becoming commodities
  27. AI news roundup: You see what you want to see
  28. How to build your first LLM evaluation
  29. Open LLMs don’t need to beat OpenAI
  30. AI must be more than a checkbox
  31. How to talk to someone who doesn't trust AI
  32. Lessons from customer evaluations of an AI product
  33. You can't build a moat with AI
  34. We need better LLM evaluations
  35. RunLLM: The custom assistant for developer tools
  36. Untangling concerns about consolidation in AI
  37. Don't build your application on a single LLM call
  38. Building on quicksand
  39. A tornado of AI news
  40. There's more to LLMs than chat
  41. LLMs are the variable interest credit card of tech debt
  42. An introduction to evaluating LLMs
  43. Prospecting in the AI Gold Rush
  44. Fine tuning is just synthetic data engineering
  45. 2024 AI Predictions

2023 17

  1. Rumors of startups’ deaths have been greatly exaggerated
  2. OpenAI will solve its privacy problems before you
  3. One LLM won't rule them all
  4. Why open-source LLMs matter
  5. MemGPT: Memory Management for LLMs
  6. OpenAI is too cheap to beat
  7. Open-source LLMs shouldn't try to win
  8. Gorilla: An LLM for Massive APIs
  9. LoRA, explained
  10. How our Test-of-Time Paper Almost Wasn’t
  11. RLHF and LLM evaluations
  12. How to Optimize Retrieval-Augmented Generation
  13. Should you fine-tune a model?
  14. The Easiest Part of LLM Applications is the LLM
  15. A deep dive on Llama Index
  16. Building your first LLM-powered application
  17. Welcome to Generating Conversation