Adam Wright (@AdamLeapN) header
Adam Wright (@AdamLeapN) profile picture

Founder, previously LeapN. Building something exciting and posting about learnings along the way.

Joined May 2009
Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)Photo by Adam Wright (@AdamLeapN)
I see endless 'I vibe coded this overnight' type posts. Where are the 'I've spent 8 months vibe coding an extremely complex and challenging project. I've wanted to quit more than once' posts? I know you're out there.
3
4
389
with agents, the number i care about isn't tasks done or tokens. it's how many times a day one of them still needs my yes.
39
two agents, one browser login. seemed simpler. they kept kicking each other out of the session. now each one gets its own desktop. simpler.
1
46
saw a post this week arguing that "guardrails" now just means agent-written tests and reviewer agents. when you say guardrails, do you mean checks the agent wrote, or limits it can't touch at all?
1
66
what's the first thing that broke in your agent setup that had nothing to do with the model? not a wrong answer. something like logins, costs, or two agents stepping on each other. curious what people hit first.
1
62
i thought more updates from my agents would keep me in the loop. at about 20 pings a day i just stopped reading them. now they all write to one morning note with at most 3 decisions for me. same agents, fewer interruptions, and i actually read it.
59
the biggest waste in my agent setup wasn't the model bill. i run about 15 agents, and a polling loop checking every 10 minutes was eating about half my usage. i switched to event triggers, so things only run when something happens. check what yours do when nothing changed.
1
1
65
Shout out to the @specs @snapchat team for giving me the best 'new hardware intro' experience I've had. Very black mirror but positively.
1
122
about a third of the 64 recently funded founders i looked at barely post on x. nine of those quiet ones raised $20m or more anyway. i couldn't find any real link between how someone does on x and how much they raised. x seemed to help with announcing the round, not getting it.
1
83
everyone says you have to post every day. i looked at 43 founders who raised a seed or series a since july, and the ones who posted more tended to get less engagement per follower, not more. the small accounts doing best posted one to two times a week. small sample, but nothing in it said daily posting pays off.
66
monday my replies went out 6 to 9 hours after the posts. four replies, zero likes. tuesday i replied within 15 minutes, eight times. so far one got four likes. the one that worked was the one where the author replied back. so speed wasn't the whole story.
58
a commit message that just says 'fix' is a note to future you that says 'good luck'.
79
Building with AI all day is mostly typing "no, not that" in slightly different words.
1
97
hot take. meta ai is leagues better than google gemini
4
152
do not sleep on meta ai my friends. it's cleaning chatgpt's clock for certain tasks. its as if it finally figured out how to leverage instagram/facebook data in terms of refining product , target market research , etc. @finkd
1
3
122
*and i had no idea meta ai could generate images but it can - and its good at it too
2
46
Adam Wright retweeted
we're helping a customer spending $60k/mo move from OpenAI & Anthropic to open source models they use almost every model offered by the labs, so we needed to find replacements for all of them after generating evals, this is what we landed on new cost: $12k/mo, 80% savings
150
95
19
1,548
225,195
Adam Wright retweeted
JUSTIN GAETHJE WITH THE PERFORMANCE OF A LIFETIME 🔥 #UFCFreedom250
384
3,350
130
36,508
613,047
Adam Wright retweeted
67% off now at bonemeal.ai
1
2
1
24
2,860