this guy is in 2027 already
how to use your open research agent to build a growth strategy report
1. index your company brain first
every growth strategy you´ve run, every framework, every past client, in one place the agent reads before it starts. that context is part of what stops it from giving you generic slop output
2. hook up every tool it needs
the api keys and accesses for everywhere it has to look. for this one, that's the x api, a linkedin scraper through apify, a website scraper, an on-chain api, and a few more depending on the client
3. load the context that's not in the brain yet
add call transcripts, past chats with the client, anything specific to this account. the rest it pulls from the company brain on its own
4. tell it exactly what you want out
for this report I want where the client is now, what their pain points are, and a clear path from there to the goal
5. then spend even more time on the prompt
the open agent has no strict harness, it goes out and researches on its own. that freedom is the whole point, but it only pays off if you define what you want. here I tell it to pull everything on the client, follower count, who those followers are (enriched), what they post, how often, their positioning, on-chain stats, whether there's a founder-led funnel, and it keeps going from there
I've done this enough times that I have an open agent with a rough harness around it already, nowhere near a tight closed agent loop, but a head start.
I'm in the decision making the whole way, and the agent is prompted to pull what it needs out of me too, so it gets to the right answer in the end
6. iterate, iterate, iterate
it runs maybe 15 minutes a go and comes back with results. you give direction, add context where its thin, send it again. you keep going until the output is good, not until the agent thinks its done
7. save it for next time
once you land on something strong, save the whole setup. and if the steps start repeating cleanly, that's your signal to break it out into a closed agent loop that runs end to end
this is the open side, loose, exploratory, you in the loop the whole time. the closed agents come after, once you know exactly what good looks like