a blog about LLMs, Agents, and AI, by Peter Sergeant
I used to write little translators for my bank exports. Then ChatGPT wrote them. Now nobody writes them: I hand an agent the file. As agents get better at following instructions, more tasks stop needing durable, task-specific software at all.
I hit enter too early on a message to Claude and it identified the fundamental design flaw in our codebase before I could describe it. Three LLMs had signed the thing off as essentially perfect. Nobody was in "are we solving the right problem" mode.
I posted my AI agent jail to Show HN. Nobody upvoted it, but within fourteen hours thirteen LLM-written sales pitches found the two email addresses on the page.
We need a better vocabulary for describing contribution in the age of generative AI. The binary of "human-made" versus "AI-generated" lacks the expressive power to say what a piece of work actually represents.
Over and over again, despite the best efforts of humans, the most effective AI systems come down to one simple idea: finding the right shaped line that fits some data points.
Existing articles often focus on a chunk as a singular concept: you split the article into paragraphs, say, and use these to feed the LLM, generate embeddings, and quote back to the user. But that's three problems, not one.
You delivered your vision quickly using one of the many excellent LLM-based coding tools, users are happy. I will dig you out of the tech-debt hole you have created.
Embeddings are a core AI concept that underpin a great deal of what we today think of as being AI. This article is going to give you an accurate and intuitive understanding of what an “embedding” is in less time than it takes to eat a (very large) bagel.
In our application, the chatbot can’t hide behind a loading spinner; users keep talking and expect it to pivot instantly. This constraint forced us to develop some lightweight techniques you can graft onto your own LLM app that serves impatient users.
While working on LLM-driven NPCs, I observed significant improvements in several areas by adding a simple component: in-memory free-text search
Don’t let an LLM make decisions or implement business logic: they suck at that.
If your team promises to deliver (or buy!) 'Agentic AI', then everyone needs to have a shared understanding of what that means; you don't want to be the one left trying to explain the mismatch to stakeholders six months later. There's no current (2025-03-30) widely accepted definition, so if you're using the term, be clear on what you mean, and if someone else is using the term, it's worth…
or, reducing hallucination and making in-generation adjustments to LLM responses