I spent too much time on X. The usual AI FOMO.
So over the last weekend, I built X Analyst.
X is still one of the best places to find interesting AI stuff early: model releases, agent experiments, small open-source projects, paper discussions, weird demos, implementation details that somebody casually drops into a thread.
But the problem is that all of this appears inside a feed. And feeds are not designed to help you think; they are designed to keep you scrolling.
So I made X Analyst.
It is open source for non-commercial use: florian-zeev/x-analyst.
§What it does
X Analyst watches X sources and linked articles, filters them through what you care about, and sends you a short daily brief by email.
The basic idea is:
- tell it what matters to you,
- let it decide what deserves to make the brief,
- give feedback on what was useful or not,
- get a better brief over time.
§Why an AI analyst
Generic AI news is fine, but generic relevance is not personal relevance.
For example, I might care about a small technical detail in an agent framework because it changes how I think about building something. Somebody else might only care if the same framework has enterprise support, SOC 2, or a specific deployment story.
Both are reasonable. But they are not the same query.
That is why I wanted something more like an analyst than a search engine. Not just "find AI content", because there is too much AI content. More like:
- which sources do I trust?
- which topics do I currently care about?
- is this actually new, or just newly packaged?
- is the linked article more important than the post?
Doing this manually is a huge time sink.
§Using Eve
I also used X Analyst as a playground to try Eve, Vercel's new agent framework.
It is really good.
The "Next.js for agents" framing makes sense to me. An agent is a directory. You can start with instructions, then add tools, skills, channels, schedules, subagents, and sandboxing as the system grows.
For X Analyst, that maps nicely to the actual shape of the product:
- behavior: find the AI-related X items I should not miss,
- tools: read posts, inspect links, store briefs, send email,
- schedule: run every morning (or whatever your favorite time to receive a brief is),
- feedback: remember what I found useful.
What I liked is that these pieces feel like natural extensions of the same system, not a pile of unrelated infrastructure.
§The larger point
I think this is a good example of where personal AI tools are actually useful.
Not as giant universal assistants, but as small bits of software that remove one recurring annoyance from your day.
For me, that annoyance was checking X "just for a minute" to see what was happening in AI, and then spending half an hour separating useful things from noise.
So I built a small analyst for that. I'll probably build some more in the future.