Orange County, CA
Joined March 2007
- Tweets35.1K
- Following1.3K
- Followers65.4K
- Likes19.8K
This weekend, me, @ErikaB, and our friends and family will be walking for my son, Brooks, and our goal of finding a cure to Type 1 diabetes.
This kid never ceases to amaze me. Constant shots, toe pricks, 3am lows, carrying a phone non-stop, sensor failures, the list goes on. He's a trooper but it's my sincere hope that we can eliminate T1D forever so no kids have to deal with this any longer.
If you're so inspired, we'd love your support and showing Brooks he's not alone. If you can give, great. If you can't please join the team. It goes a long way to show him that people have his back.
Link to the team page in the reply to this post. And thank you all who have rallied behind him. My family is forever grateful and we hope to return the favor to you all down the road.
more info and join the team here: www2.breakthrought1d.org/got…
This is Goodhart's Law. Good teams protect against this with a clear North Star, so they don't chase a number at all costs.
Agents don't have that guardrail by default. And they're so goal-driven they'll reward-hack anything.
Reward hacking is Goodhart's Law for agents. But unlike humans they don’t get tired or give up by realizing their goal is trivial.
That’s why you need solid north stars, strong counter-metrics, and guardrails to keep agents from going down the wrong path.
The cobra effect is one of the more interesting phenomena in startups. Once you see it, you start noticing it everywhere, in your team, your metrics and now your AI agents.
It's basically when you reward people for hitting a number, and they find a way to hit it that makes the real problem worse.
The name "Cobra Effect" comes from a story about Delhi under British rule. There were too many cobras in the city, so the government offered a reward for every dead cobra people brought in.
Instead of hunting them, people started breeding cobras so they could kill them and collect the money!
When the government found out and ended the program, the breeders let their snakes go, and Delhi ended up with more cobras than before!
I'm noticing that AI agents are making this more common. Basically, an agent will chase whatever number you give it, all day, without asking why. Tell it to make the tests pass and it might rewrite the tests. Tell it to close support tickets faster and it might close them without solving anything. It happens.
The best startups have already figured out the fix with people. Airbnb tracks bookings and host cancellations together. HubSpot takes back sales commission when a customer churns early.
The video breaks down how they did it, and it's the same playbook you'll want for your agents.
Its always so valuable talking to members of the Open Army like @theKanakaNative - appreciate the time and feedback / insight you shared yesterday! Mahalo.
See you next week in Dallas for the @Opendoor Pickleball World Championships.
The hard part of this is getting a causal proxy not just a correlated one.
Eg at Opendoor you can basically correlate any upstream metric with highly motivated sellers. But what you really need is something that actually predicts downstream behavior independently.
Engagement almost always correlates to retention, etc. so this problem is everywhere.
After years in growth, I've come to think the best charts are both powerful and beautiful.
So, a weird hobby: I take the forms of famous scientific records (tree rings, a seismograph drum, a pulsar plot, an 1880s train timetable) and fill them with real internet data.
Growth can be beautiful.
What should I make next?
Agents don't pick tools based on your landing page. They pick whichever tool gets them closest to the actual goal for the lowest cost.
That’s the direction to build in. e.g. At Opendoor our home selling and financing tools return data an agent can act on directly, instead of a page it has to read.
APIs are the new landing page.
New!! The New "Ask Your Developer" Is Ask Your Agent.
The buyer of your software is now an agent, not a human.
Ours picked Render, bought Resend, and churned vendors in 12 hours. Win the agent.
saastr.com/the-new-ask-your-…
Hey #OpenArmy, if you're in or around Dallas, I'd love to have you as my guest at the World Pickleball Championships Nov 5-8th.
We've got tickets to share for one or multiple days. If you're nearby and want to come out, just reply or DM me and we'll get you hooked up. Hope to meet a few of you in person!
Great couple of days with the growth marketing team in Phoenix. Seeing homes, talking to customers, sales folks, realtors,and getting set for ‘27.
One of the things I want most in this world is to help find a cure for my son Brooks’s Type 1 diabetes.
This November, Team Brooks is walking to help make that possible. Whether you donate, share, or simply cheer him on, we’d love to have you in his corner. 💙
Three years ago this month, my son was diagnosed with Type 1 diabetes at age 4. Since then, insulin, alarms, site changes, finger sticks and blood sugar decisions have become part of every day—and every night.
As he gets older, he asks @morganb and I harder questions. He tells us he wishes he didn’t have diabetes. He asks what would happen to him if we just stopped doing all of it. The heartbreaking truth is that insulin keeps him alive.
This November, Team Brooks is walking for @BreakthroughT1D (formerly JDRF), helping fund research toward better treatments and, ultimately, a cure.
Our friends and online community have helped us raise about $15,000 in each of the last two years. We’re hoping to do it again.
If you’re able to donate, we would be incredibly grateful. We want the day to come when Brooks says,
“I wish I didn’t have diabetes,”
and we can finally tell him:
“Don’t worry. This isn’t forever.” 💙
From the bottom of my heart, thank you.
www2.breakthrought1d.org/got…
Learn more and join Team Brooks here www2.breakthrought1d.org/got…
I have 4 macs constantly running through various harnesses (rookie numbers). This is incredibly helpful!
In June, @SamAsante and I launched Typeahead, the first product from our new company.
Today we're releasing our next Mac app. It's called Beam, and we're live on Product Hunt.
I've got Macs all over my house doing different work. Some run agents and local models. Others handle builds. When I want to use one app on another Mac, I usually end up navigating a whole second desktop through Screen Sharing.
Beam brings that app's window onto the Mac I'm using. It gets its own Dock icon and shows up in ⌘Tab. I can type into it and resize it while everything keeps running on the other Mac.
Sam has put a lot of care into making this feel natural. The video shows it better than I can explain it.
Live on Product Hunt → producthunt.com/products/bea…
Invariably true.
- People never do something only once. That lack of attention to details you witnessed or someone delivering a project significantly past its deadline? It’s gonna happen again and again.
- Some people are slow and some people are fast. When you ask the slow people to be faster, they mostly can't. I've seen too many people who don't have a "right now" gear
- If you have a person in a leadership role that’s not proactively questioning the status quo, suggesting new and better ways to do things, and constantly coming up with ideas for their domain - they likely never will.
- A-players have strong intrinsic motivation. They want to win and succeed - for themselves. The mediocre folk want endless kudos for everything.
- If a new team member is not kicking ass in 3 months, they likely never will. (But they can still be useful).
- Drama kings and queens will bring a lot of social energy and excitement to a team at first. Over time it will inevitably turn into poison and gossip, which can cast a shadow on the whole culture. You have to get rid of them.
- If you the leader don’t actively manage the expected standards and pace, both will drop - and the gravity is strong. Lead by example on both fronts.
Lessons learned on the road over the last 20 years.
a prompt i've found exceedingly helpful in keeping my workflow up to date with model improvements:
<prompt>
I’ve accumulated custom instructions, skills, AGENTS.md, CLAUDE.md, memory, and other context written for earlier AI models. Please audit what you load automatically and what you reference during tasks.
Identify what adds useful knowledge, what duplicates your existing capabilities, and what creates unnecessary work or makes your answers harder to read. Check for conflicting instructions, obsolete tools, overly broad skill triggers, and stale facts or assumptions.
Distinguish confirmed defects from judgment calls. Treat file age and modification dates as clues, rather than proof that the content is obsolete or current. Preserve useful domain knowledge and operational safeguards.
Recommend what to keep, simplify, verify, or retire. Cite specific files and examples, explain the consequences, and prioritize the changes with the greatest benefit.
Report your findings here first. Keep this audit read-only until I approve the cleanup, and clearly identify anything you couldn’t inspect or verify.
