Dispatch · Models ·
What builders should watch after the LLM hype cycle
Chatbots were the headline. Builders should watch tools, reliability, cost, and where models sit inside a real product.
For a few years, “AI” in public mostly meant chat. Type a prompt, get an answer. That made demos easy and products blurry.
If you ship software, the useful question is narrower: what can a model do inside a real stack without wrecking cost, trust, or maintainability?
Chat is table stakes
A chat box is useful. It is rarely a moat. Customers pay for outcomes — a finished workflow, a safer process, a faster ops loop — not for the novelty of talking to a model.
Where things get interesting
The next layer is tools and agents: models that call APIs, edit files, schedule jobs, or move through multi-step work. That is also where failure modes multiply. A wrong paragraph in chat is annoying. A wrong write to production is expensive.
Builder takeaway
Treat the model like a component. You still own the problem definition, the architecture, the evals, and the decision to ship. AI can accelerate drafting and research. It does not replace engineering judgment.
Sources
-
Yann LeCun on What Comes After LLMs (Unsupervised Learning)
https://www.youtube.com/watch?v=ngBraLDqzdI
-
Yann LeCun interview: LLMs nearly obsolete (Newsweek)
https://www.newsweek.com/nw-ai/ai-impact-interview-yann-lecun-artificial-intelligence-2054237
-
A toddler beats ChatGPT: Yann LeCun world-model bet (The Next Web)
https://thenextweb.com/news/yann-lecun-ami-world-models-raise-summit-llms