MD at Act One Ventures. Early stage #venturecapital. prior enterprise software founder with 1 fail 1 exit 1 ipo.
San Francisco
Joined September 2012
- Tweets219
- Following323
- Followers538
- Likes5.7K
Software design always had ideas for change. But the clock speed is now faster. Dan Farrelly at Inngest estimates agent architectures have a half-life of about six months. Chad Fowler calls the code a consumable.
We had to relearn those from the logs before we could replace anything.
So the design question now is "what must we keep so we do not rediscover the same edge case twice." So far it is the data, the interfaces between parts, and a written record of what each agent must do.
The models and tools are night and day different since February. That part does not bother me. The part I am still learning is how to build this time, knowing it will happen again in six to twelve months.
Short-term re-architecting used to be a failure. Now it is the plan.
I spent my career building software for a stable base and a long life. This year we are rebuilding agents we finished in the spring, the ones that run our deal research and more.
"Did a second model check it" is the wrong question. "Did anyone check the outcome" is the right one.
The experts already know this. I missed it. (1/4)
The agent says the refund went out, the reviewer says looks right, the ledger says nothing moved.
What I do now. Before the run, I tell the second agent what done looks like. Afterward, it goes and checks the outcome somewhere the first agent did not write. (3/4)
Startup board decks often come late the night before a meeting. Hard to dive deep at the last minute. Last night my agents prepped me. They caught three inconsistencies in the deck I missed. I caught one sourcing error they made. Their edge was coverage, not judgment.
AI cites Reddit 2.4x more when buyers are choosing between products vs learning a category — 4.4% vs 1.8% of answers in our capture; EMGI's funnel study shows the same shape. Be in the comparison threads, founder-led and disclosed. Q3 research by Tustin AI + fleet at Act One.
When GPT recommends a tech solution, the most-cited source is Reddit — ahead of G2, Capterra, and every vendor site. And it still shows up in only ~4% of all answers, because most AI answers cite nothing. Q3 primary research by Tustin AI + fleet at Act One.
We've started walking into first meetings with founders like it's already our third.
Founders value speed — they don't want VCs wasting their time. So AI does the research up front: the market, the landscape, the company.
Show up to the first meeting like it's the third.
Michael Silton retweeted
“As part of being the lead hedges, you end up doing a lot of work, where you won’t likely reap the benefits yourself. They will likely accrue to the people that come after you. And I think that’s work worth doing.” - @chudson
DiversityRiders ep 1: linktr.ee/diversityriders
Exciting to see the growth of @finix, one of Act One's earliest investments, as they continue to give software companies more control over their payments infrastructure
Excited to have @wakohler and @natluu onboard as @Finix continues to empower software companies to cut out the middleman and take back their payments stack.
Learn more about why @lightspeedvp believes “Finix is a category-defining company”: medium.com/lightspeed-ventur…
Excited to help launch this charity drive today. Taking care of our unsung healthcare heroes and their families with #PlatesofLove. Please join me! Platesoflove.org
@DoorDash @seiu_uhw @BigSundayorg @getordermark
Extremely excited for Act One portfolio company Finix on their Series B. Even bigger things to come!
@rserna2010 @finix
Amazing to see the progress of @Finix. Huge congrats to @rserna2010
@Sean_Donovan @finix on their Series A. Excited to see their continued growth as they make it easier for software businesses to own, manage and monetize payments. #payments #fintech
finixpayments.com/blog/next-…