start of a journey....
built a app before - the app actually worked. still failed. not an engineering problem, a distribution and validation problem. built in a vacuum for months and found out too late that nobody was actually looking for it.
so here's the correction, done in public, on the record:
90 days. learning ai agent frameworks (langgraph + mcp), documenting the real process, and validating an actual business problem with real people before building anything meant to last.
the plan, roughly:
weeks 1-3: learn the fundamentals, build small things, get the posting habit locked in
weeks 4-6: talk to 20-30 real people in one industry. not pitching, just listening for what actually hurts
weeks 7-9: build the smallest possible version of whatever comes out of that
weeks 10-12: try to get someone to actually pay for it
i don't know the final product yet. that's kind of the point - finding out in the open instead of guessing alone in a room.
day 1. let's see how this actually goes.
been thinking about what to actually build with agents - most "is this agent safe" debate treats it like a yes/no question.
the real dividing line is way more specific: does it just draft/flag/suggest, or does it actually execute something irreversible on its own. an agent that drafts an email for you to send is a completely different risk category than one that sends it
almost every genuinely useful, low-liability thing i can think of building fits inside "human approves before anything real happens." the risky stuff is specifically what falls outside that line
Planning to build a portfolio site to share and showcase all the agents that i've been building - a real browsable place instead of them being scattered across posts
stay tuned for the design and build logs
been thinking about something since building the drafter agent - most people's mental model of "ai assistant" is still one-shot: ask, get answer, done
the more interesting stuff happens when it works more like a real collaborator - draft, review, revise, draft again - closer to working with a person than querying a search engine
feels like a lot of what "ai agents" actually means in practice is just building that back-and-forth loop properly
Letโs connect ๐ค
Drop what youโre building below.
Apps.
SaaS.
AI.
Startups.
Side projects.
I wanna connect with more people actually building stuff