
Dad | AWS serverless hero | farmer | web dev | enterprise architect | product @momentohq
McKinney, TX
Joined February 2012
- Tweets4.3K
- Following1.1K
- Followers5.3K
- Likes12.7K
It took about 10 turns, but I was able to get the @HomeDepot chatbot to write some code for me to determine an objective score for rating power drills 🤣
Remember people, guardrails aren't perfect! Where there's a will, there's a way.
Adding a vector store will make your agent's memory dramatically better, right?
Not necessarily.
I recently did some benchmarks on retrieval methods and am genuinely surprised at the results because they contradict every talk I've seen on production RAG 🧵
My main takeaway is to not add AI components because that's what articles on the internet or speakers at conferences say to do.
Benchmark them.
Every reranker, model call, and retrieval stage needs to justify its latency and complexity.
Full benchmark, thoughts, and code:
readysetcloud.io/agent-memor…
Sponsored by @Oracle and @oracledevelopers. Opinions are my own.
I learned last week you can run code in the chrome debugger in the console tab. For over a decade I've thought that was a read-only tab
The amount of wasted tokens from repeated tool calls, metadata lookups, and semantically-equivalent messages in AI agents today is staggering.
In 2025, I demonstrated how to build idempotent tools and how to design cacheable APIs for agent use. But that wasn't enough. 🧵
My friends at @betterdbhq recently released two packages on npm and PyPI that make this dead-simple to add to your agent logic with only a few lines of code.
I wrote up why caching is more important than ever and how you can add this to your apps in an afternoon. 👇
rdyset.click/c/ktoSJI/
I built a local AI agent with its own identity and thought I was doing security right.
Boy, was I wrong.
Identity is one layer. It doesn't close the four attack vectors that come with running an agent on your machine. 🧵
These aren't identity problems. The best identity model in the world doesn't close any of them.
But what does help is moving the agent off your machine entirely. Ephemeral runtime. Fresh environment per task. No accumulated state except what you give it.
I wrote the long version with real scenarios for each vector, and where Teleport Beams helps address these issues for a truly secure agent environment.
Thanks to @GoTeleport for sponsoring this post, it was really fun to write.
readysetcloud.io/blog/allen.…

