Co-founder of @RemotelyWorksHQ. Previously co-founder @Olapic (acq by Monotype). Alum @bcg @columbia
NYC
Joined March 2007
- Tweets4.4K
- Following1.2K
- Followers1.3K
- Likes2.5K
Pinned Tweet
it's been 48h since i started using @lovable
now i'm the proud owner of 4 non-functioning web apps.
Growth at a startup directly correlates with proving ROI.
And if you tie it to revenue growth it will generally grow faster than reducing costs.
Prosper is a great example of a startup connecting AI to value creation.
I spend my day speaking with leaders of healthcare practices.
2027 is going to be a very tough year.
CMS reimbursements are down 9% in specialties like Dermatology, and 84% of healthcare practices report an increase in their operational cost.
Healthcare practices are feeling an incredible squeeze, and many will go out of business.
But leaders have more control than they think.
People think about AI primarily as a way of cutting costs. But AI is the opportunity for them to take over control and fight the growing pressure on their ability to provide quality care.
We quantified exactly how much control practice leaders can count on by introducing AI into their practices in four ways.
Today, at Prosper we’re publishing the 2026 Prosper AI Growth Report, showing how AI can drive up practice revenue by $125K/year (11% increase) per physician.
You can read more about our report in the link below.
Long-horizon agents may remake today’s internet more than another model release.
Noah Shinn / Instinct (Colossus): once people have an agent that pursues goals over days — text, call, email, click — most product design assumptions look dated.
Services were built for humans doing the work in the UI. If everyone has an agent, a lot of those interfaces become plumbing. The agent doesn’t need your settings page. It needs a clear path to finish the job.
And that might be the extinction event for friction-as-business-model: cancel mazes, phone trees, “are you sure?” screens that monetize how busy and slightly lazy humans are. An agent doesn’t get tired of the NYT-style retention gauntlet. It just cancels.
The internet optimized for human patience starts looking temporary. Rebuild for agents as first-class users — or keep betting that people will give up halfway.
Designing for the agent yet, or still for the tired human?
Old: we decide what we do → ask AI for help
New: AI decides what we do → AI asks us for help
Pat Grady’s AGI-is-here take, rephrased: intelligence already plans better than most people. So AI starts taking the steering wheel on execution; humans keep intent, goals, strategy.
Same shape as self-driving. Once it’s safer than the average driver, refusing the wheel isn’t careful — it’s stubborn (and, from a shareholder seat, irresponsible).
Seeing early weird versions already (AI with its own YouTube / personality going to build an audience on its own). Software, marketing, sales will follow the same flip.
Are you still in old mode, or have you handed the wheel on anything and only jumped in when the AI asked?
As predicted.
AI, at its core, transforms how the corporation processes information.
The design of how you implement it and the iterative knowledge you derive from it cannot/should not be outsourced to an external FDE (who siphons the knowledge back to the software provider).
A big trend with most enterprises that I talk to is deploying internal FDEs into departments in their organization to help bridge the capabilities of AI into their underlying workflows.
This requires a high level of technical expertise, an understanding of AI, and the ability to understand the workflows and processes that the enterprise is trying to automate. There’s no shortcut to getting automation without this combination of skills (either in one person or multiple).
What’s exciting is this is an entirely new function in most enterprises, which is going to create a ton of new roles in the economy.
“Every generation of technology creates jobs that didn't exist before it, jobs that arise from an immediate need in an emergent industry. The Automation Engineer is a key example of those jobs for this generation. Nobody has ten years of experience doing this yet, because ten years ago the tools that made it possible didn't exist (it didn’t even exist two years ago).”
If you have software skills and are diving into AI, this is an area to go deep on.
It's interesting to see the counterargument to the @theallinpod point of view.
Either you believe AI is just too dumb to be dangerous (we've all been frustrated by that) and think the demand for regulation is a regulatory capture strategy...
... or that belief is AI's way to have us lower our guard, and eventually get rid of all of us.
Six months ago i wrote that i use one model for thinking and a cheaper one for doing. A model launched last week that does not write at all, and the cheap part is not the interesting part.
Jev, from TypeSafe. You hand it data and a list of typed questions and it hands back decisions, in about a tenth of a second, at $0.042 per million input tokens and no charge for output.
Until now, regularly running a model across a huge dataset was too slow and too expensive. So you cut the dataset down before you sent it, and built the workflow around the smaller pile.
That was not the best way to work. It was a hack to manage time and cost. But a hack does not announce itself as a hack once it is in production. It just becomes how the thing is done.
Now, for this kind of decision, you can run the model across the whole dataset. So the question is not how much cheaper your current workflow gets. It is which parts of it only existed because of the old price.
What hack in your process have you stopped noticing?
Great clip. Wrong take.
The fact you don't enjoy doing things, doesn't mean you don't need to do things. Anyone with kids needs to schedule a dentist, pediatrician, deal with insurance reimbursements, deal with afterschool activities, find things to do on a weekend...
I know plenty of parents that are going to jump head first at the promise of the truly autonomous assistant that takes care of all these family tasks.
So many things in technology follow this story arc.
Crazy innovator uncovers the primitive concept, but it is clunky and unpolished.
And then someone, often times an incumbent with an eye for spotting great products, just rebuilds it better.
we built muse from scratch, but it is definitely heavily inspired as a product by openclaw. After I used openclaw in january I bought hundreds of mac minis for the MSL team and lots of us fell in love with using openclaw (and other personal agents). @steipete is a genius and his harness was pioneering from the jump. I think a lot of people were inspired by it. our goal with muse was to build something like openclaw that we could make safe and secure and easy to use and scale to billions of people.
Today my Instinct struggled to complete a form to submit a refund. I stepped in and found out that the instructions were asking for something the form was not allowing (submit two photos, but the form allowed for only one attachment).
Sort of anti-agentic captcha.
In the end, it seems a poorly designed UI will be the last bastion AI can conquer. The final level will be Microsoft's 365 license management portal.
I despres diran q son els expats els que es carreguen la carniceria de barri.
🔴 Barcelona tindrà una xarxa de comerços assequibles semblant als súpers públics de Nova York
beteve.cat/politica/barcelon…
The phone plan for agents was an obvious idea when i started playing with openclaw.
I wonder who is building this in europe.
Give your Grok Bot a $3 phone plan
To celebrate Grok @bot's 3 days of livestreaming, your first month of the Agent Phone Plan is $3. Available for 3 days only.
Setup now takes under a minute. Start with our Grok Bot template:
x.ai/bot/4BmzjFu_i3mVd58irr1…
This phone plan give Grok Bot:
- A dedicated phone number
- the ability to send and receive texts
- a voice to make inbound and outbound calls
“That app/saas should have been a skill” is my new favorite framework.
Whats your example? Mine: Lovable.
A bad shortlist never shows up on your invoice. It shows up on the calendars of the most expensive engineers on your team.
Vendors move fast on the part they control. Sourcing is quick and countable. The judgment call is slow, and it isn't the thing being measured. So it is easy to look fast. The screening doesn't disappear. It moves downstream, into first-round calls your staff engineer sits through and then writes off.
A fast shortlist can still be an expensive one.
At @remotelyworkshq the median shortlist takes 48 hours, and 80% of the people on it get interviewed. We are fast. We get there with AI on a 7,000+ network a person cannot comb, and humans deciding who is worth putting in front of you.
I care about the 80% because it means most of the list is already worth a conversation. Your staff engineer is interviewing people they actually wanted to meet.
80% interviewed is not 80% hired, and i wouldn't want it read that way. Whether someone is right for your team is your call, and it should stay your call.