noashavit

A framework to build self-improving persona-based agents. Comes with 3 skills, agent template, and built-in feedback loop.

Governance, context and cost management are fundamental. Without these, most ENT agentic deployments are blocked

Interesting how close sonnet 5 comes to opus 4.8

looks super useful- thanks for sharing!

Uber, Microsoft and now Meta. All tokenmaxing to the max on Claude

The more AI is used the more we crave the human connection

Local models on consumer electronics are surpassingly good for certain tasks, like summarization and AI overviews

It's quite obvious that things are highly subsidized now. What will happen when subscriptions 10x in a year? That's that danger for the enterprise. I don't think the average enterprise will allow users to run local LLMs tbh. Just my POV feel free to tell me that I'm wrong!

I don't think we're at an aging horse moment, but a hurt one. One that can run fast sometimes but limps to the finish other times

I hear you lose access to Claude Design the minute you cancel you subscription.

Wouldn’t GDPR require they opt-in rather than out?

Interesting move in to fully managed service by Anthropic.

Anyone here remember dos? The og CLI bases OS

If I was a lawyer I would worry about auto note takers too. So easy to view something out of context as something totally different from what actually happened. Just my 2 cents

Tbh I don’t know if any careers are lifetime these days. Maybe plumbers have job security. GPUs need cooling…

I don’t think low quality agentic support is the future. But agentic support with escalation routes to humans makes sense to me.

Support agents are a brand’s first line. It hope they invest is good agentic support to protect their brand. Time will tell I guess.

Gitbook gets the job done

Gemini for recent search and google workspace automation

Perplexity for deep research

Claude Opus for coding, Sonnet for writing

Gemma4 for local AI overviews and analysis

Qwen coder for local prototyping

Lakebase is based on then Neon, that is why it was acquired. These are the performance gains from that underlying tech

Gemma4 is a huge improvement and it's fast. Qwen 3.5 really slows down my machine though. LMK if there is a better model to use for the code assist aspect- performance wise

Relying on external APIs network failure points and unavoidable latency from the round trips. There is also the AI API rate limits that come into play. We might find that for critical workflows, local compute is the only reliable architecture.

The race for unstructured data continues. It feels like everyone is trying to crack unstructured data extraction with the underlying goal of ultimately using AI to classify and tag insights from unstructured data to create a structured data/graphs for agents to consume and traverse.

What you built, summarized for you.

This looks helpful. Context is the main issue agents face. How does this fix issues with messy data though? Like in the real world when you have 2 reports for the same metric with different results. Will the agent have enough context to provide the same, accurate, result consistently?

People in SF need this. Too many cracked open Macs running agents have been spotted outside public bathrooms haha ;-)

Looks interesting! Your demo vid in the README didn’t load for me btw.

I cannot agree with this more. This is a turning point for many. Opportunity is the name of the game if we just view it that way.

"somewhere in your org, there are ~five people who are going to deliver you 100x the amount of value you ever thought possible. Your first job is to figure out who these five people are (no matter how junior they are on paper!), explain the urgency of the situation with them, and give them the career opportunity of a lifetime to rebuild the company with you."

I would do 1 and 3 in tandem until you have confidence in the agent’s output across different use cases (personas, channels).

Start on a small limited scope. It should be a something you can do in your sleep with data you know and understand. Define a narrow scope and thorough guidance for the agent. Then have it create a plan, draft the outreach, and present to you for approval before executing. You can have it send you an email or slack you with everything.

Take the extra time until you have trust in the output, because real world messiness breaks most agents. Even those that worked well as prototypes.

It’s not just cost per seat. It’s lock in, eroding skills, latencies. I worry about this a lot. There are companies that rely on Claude or Cursor in a way that is not easy to rip out, even if rates 10x.

How is this different from using local models with Ollama?