Artem Petakov

@akothari @NotionHQ I much prefer the Coda idea where every document is like a whole small Notion. So easy conceptually.

Do you absolutely hate @NotionHQ? If so, I would love to chat with you (feel free to reply or DM!)

@howard @roam If I run 20 sessions in parallel, does it upgrade me to a corner office?

Now in @roam, when you're Claude Coding, your office heats up and expands as Claude is writing code. This is tonight's easter egg in your favorite Virtual Office Platform!

@ManOfFocus_ Candid but kind.

As a man, what is your personal philosophy? (In 3 words or less)

Really great advice here, @howard -- I've been a fan of your PMF since "Just a tip" ~25 years ago! Agree with almost every word of this, and I hope lots of people read it. Just two things to add: 1) I like LTV/CAC as one way to measure PMF. Of course there are many other related metrics, but I find LTV/CAC gets at how much people like it and how quickly can you scale. 2) Relatedly, I think the growth doesn't have to be organic to indicate PMF. Lots of businesses (from Noom to Expedia) scale with a lot of paid spend... just LTV/CAC has to be right. Some products are viral/organic yes, but that's only one way to fly.

@jasonlk What are some top agents in each space at the moment? Would love to hear your faves.

Ok the investor updates for February are rolling in. Let me be crystal clear: the ones that have the best AI Agent in their space — for real — have jaw dropping demand and growth. Jaw dropping. Unprecedented. Customers are often being turned away due to not enough FDEs, etc. And the rest aren’t growing all that much. There’s not much in the middle.

@nikillinit And the more they charge for it, the better it will work. Win-win!

I’d like to sign a form at the doctor that says “I’m okay with you giving me a placebo because I think the placebo effect is actually a really effective intervention. But I can’t know you’re doing that, so don’t tell me”

@ExaAILabs Very cool. Just to set the record straight, Google did build Google Instant in 2008, and it was quite similar. But people didn’t appreciate the extra speed. Now that AI is here, the extra speed is helpful again. developers.google.com/search/blog/20…

Introducing Exa Instant: the first sub-200ms search engine. Faster than Google, it's custom built to power realtime AI products like chat and voice.

When your website headline tries to be too clever... remember this airport ad [unmodified, not AI]. Simple works best.

@nikillinit Note also how they solved collecting outcomes measures — they get paid to collect them (attending a session at 5% WL is reimbursed, which creates an incentive to measure).

@nikillinit Don’t know much about ACO ACCESS, but in response to your #1, I think the best is to pay for both inputs and outcomes. So some reasonable payments for inputs, and then a larger payment for outputs. I think the MDPP gets this directionally right for diabetes prevention.

I think the conceptually the new ACO ACCESS model is cool - I'm not an expert here so would love to hear other peoples thoughts. Quick notes: 1) There's alway been a debate about whether we should pay for "inputs" in value-based care that we think will lead to good outcomes or pay for outcomes themselves, even if some of the outcomes can go wrong if the provider does all the right things. This seems to put more of an emphasis on outcomes and a flexibility on inputs 2) At a quick glance though, I think the most interesting part will be what happens when the measurements are collected via patient reported outcome measures. In organization I've talked to or worked out, collecting patient reported outcomes has been a disaster to implement. 3) It is interesting that pseudo-forced interoperability between PCPs and referring clinicians seems to be a part of this program. The pay will have to overcome the friction/lack of willingness to do this, but would be interesting if it does! 4) "To further support collaboration, PCPs and referring clinicians may bill a co-management payment (without beneficiary cost-sharing) for reviewing updates and documenting related care-coordination actions, such as medication adjustments or problem-list updates." - this feels like it's going to be abused and then clamped down on

@dhh 28 days later.

Today.

To each their own, but I like dedicating myself to a problem for a long time, and I’ve seen a lot of people give up too early on problems. Gates spent about 25 years on “a computer on every desk running Microsoft software”. 30 years for Bezos may not be a lifetime, but not a quick flip either. Elon is Elon.

The “find a problem that you can dedicate your life to” is so fucking contrived, I’m bored of hearing it. - Elon is solving 3-4 separate problems. - Bezos worked on Amazon for 30 years before moving onto other stuff. - Larry and Sergei quickly moved on from organizing the world’s information to everything else that Google did - gates started making OSes before moving to Xbox and everything else Just work on stuff in a space that you’re interested in and let everything else fall into place

@FintechKristen Great post! Curious what you mean by “Direct Escalation” if you don’t mind sharing.

From my humble seat as a lowly PM: - Alignment on what we were doing. A single list of what the priorities are from c-suite. - Managers as ICs. Less emphasis on coordination, more on building. - Rituals that show, not tell. Demo code every week. Including VERY rough living code to @jack - Killing consensus in favor of speed with DRIs. - Direct escalation. Disagree and commit. Make a decision within 48 hours. - Unreasonableness on behalf of customers and experience. Intolerance of process that doesn’t add value. - Hunger. Ambition. And a huge chip on our shoulder.

@chamath I’m curious if any 89-year-olds on X could speak up here if they would or wouldn’t make the choice if they had a time machine

Maybe you will live to 92 vs 89 but are you happier now?

@nikillinit Imo it’s 10-150, based on my own lived experience at Noom and also based on Rule of 150 (similar to Dunbar number). stevecorn.com/2009/02/11/rul… I plan to eventually test out the solution Gore Assoc. had for this, which is creating a new org every time size grew above 150.

IMO the most fun stage of startup is 10-100 people Below that you feel the existential dread of whether the company will exist and you'll have a job, and above that is where you start feeling the burden of process But in that sweet spot is the zone of still moving fast, feeling like you can actually impact the business, and you know each person at the company to bond with

@DeryaTR_ Sadly this vision won’t come true for a while because most restaurants don’t want you to know what they put in their food — eg how much oil/butter, how fatty meat is, how much sugar in sauce, etc. We tried this at @noom, and restaurants resisted. Need consumer protests!

Soon: “Hey ChatGPT, order food from DoorDash that you know I would like tonight. Make sure it’s under 700 calories and nutritionally balanced.” ☺️

@snwy_me One easy answer for people who claims this effect: think about all the things you talked about in the last week that you did NOT get ads about! Probably you talked about 100 things and you got an ad on one or two. Don’t you think it would be more if it was listening?

i tried this once after getting into an argument with my parents about whether your phone listens or not (this was in 2020 so may be different now) two iphone 6s’s, both wiped and connected to two different wifi networks w/ independent VPNs said keywords to one and different ones to another, never in earshot of each other, with multiple apps (insta, facebook, twitter, etc.) open with their own accounts and basic usage and the ads were never what i said they were basically always location-based/pretty basic things my running theory is that whoever thinks their phone is listening is just predictable enough that it knows, not because it heard them

I’ve believed VBC is quite broken for the past 15 years, so at least to me it’s not a short-term “trough”. I answered above that imo FFS & strict audits are better. VBCs biggest failure is that prevention has a very long term return time horizon… and most CEOs/shareholders don’t take a long enough view. This is an addition to people switching systems and attribution challenges.

It feels like we might be in the “trough of disillusionment” for VBC. At least amongst the cognoscenti. I’ve tried to zoom out and ask what would be a better answer to the implied questions of domestic healthcare. Perhaps I am too small in my thinking - or blinded by my investment positioning - but VBC does seem like the right strategy from where I sit. What do you think?

my takeaway from our healthcare data conference - the healthcare data people seem to be the most jaded about value-based care Some of the things I heard - Trying to do things that are critical for value-based care like attribution is basically impossible, so most arrangements are based on best guesses that none of the data people seem to feel very confident about - For most value-based care programs, the lion's share of money saved or made comes from documentation fiddling (e.g. risk adjustment) vs. actually changing the care model - The lag in data basically makes it impossible to actually forecast properly or intervene with patients at critical times you could actually help them Now I don't think most of the healthcare data peeps think fee-for-service is way better, I just think the optimism around value-based care is much more muted (at least in its current form)

While both systems can be gamed, I continue to believe that FFS + strict audits to prevent unnecessary work is a superior model. Especially so when it comes to prevention (so severely lacking!), where value takes many years to realize. I also like fee-for-outcome for some things (get paid if patient loses 5% of weight and keeps it off for a year).

@IBD_ECarson We don't need lidar, why should a car? We don't need wheels to move -- do perfectly fine with legs. So of course a car shouldn't need wheels either.

Vision, lidar and radar disagreeing isn't whether they like Cheez-Its -- it means that at least one sensor reading "sees" danger and at least one does not. If you're only using one sensor, there will be times when it's tragically wrong. Hard to go above L2 with that.