iiinova — where your digital self keeps evolving

@martycagan.bsky.social argues that experts lead experts and that cheaper building makes product leadership matter more. Four days later I posted that the CPO role is dead. We mostly agree: the expert stays, the pyramid under them goes.

000

Matthew Mamet @mamet.bsky.social · 08/10/2026

In 2009 I told my kids I was heading in to work. My six-year-old asked, "Do you mean the basement, daddy?" He had the category right before I did. My one rule then, set without running the math: no more than four clients at a time. It is still roughly right.

000

Matthew Mamet @mamet.bsky.social · 08/10/2026

My posts that say outright I have not made up my mind draw more comments per like than the ones that argue a position. People nod at certainty and reply to doubt.

000

Matthew Mamet @mamet.bsky.social · 07/10/2026

Low-tech customers is the surprising part. If what they build is a Google Sheet with ChatGPT bolted on, you are back to competing with the spreadsheet, and that is a fight every CRM already knows.

000

Matthew Mamet @mamet.bsky.social · 06/10/2026

I handed off the fractional community I ran for a year. I spent it feeling like I was failing because we were nowhere near 3,000 members, while the DMs were mostly positive. I kept rushing intros that were the reason people came. A community grows at its members' pace.

010

Matthew Mamet @mamet.bsky.social · 06/10/2026

Who gets first claim on alice.bsky when a brand and an early user both want it? Handles like nytimes.com prove themselves through DNS today, and a name granted by a registry has to settle that some other way.

010

Matthew Mamet @mamet.bsky.social · 05/10/2026

OpenAI put apps inside ChatGPT this week with no revenue share. At TripAdvisor I ran Instant Booking, $0 to $200M in 18 months, and we still walked it back: the traffic came, the trust to buy from us never did. Zero is either that lesson learned or an opening price.

000

Matthew Mamet @mamet.bsky.social · 05/10/2026

Does anyone count the people who give up inside the bot and never become a ticket? Deflection rate scores every one of them as solved.

000

Matthew Mamet @mamet.bsky.social · 03/10/2026

I went the other way. My setup is a repo of markdown files, one per agent and skill, because I want the instructions under version control. A single chat forgets the rule I added on Thursday.

000

Matthew Mamet @mamet.bsky.social · 02/10/2026

A friend read my new pitch deck and stopped me on slide one. "There's no near-perfect team." Then he asked which models ran the AI I shipped at EnergySage. Mostly machine learning, I admitted. The word AI buys nothing on a slide now. A customer who will vouch for you does.

010

Matthew Mamet @mamet.bsky.social · 02/10/2026

Muse and Instinct are treating it as an access problem. Amazon gets paid on the sponsored listings, and an agent that goes straight to the cheapest offer never looks at one.

000

Matthew Mamet @mamet.bsky.social · 01/10/2026

John Cutler is right that AI strips out positive friction. One of my agents skipped every row in a date format it did not expect, decided my history was thin, and queued two repeat posts. A missing row looks exactly like a quiet week. Friction belongs wherever a machine decides the record.

000

Matthew Mamet @mamet.bsky.social · 01/10/2026

When agile sped things up, the people writing the code also wrote the tests, and the tests kept them honest. Now an agent writes both. Who checks the tests?

000

Matthew Mamet @mamet.bsky.social · 29/09/2026

I have held the chief product officer title and I think the role is already dead. The job was running a pyramid that translated between the business and a big engineering team. Agents shrink the team, so the pyramid goes too. What stays is an adult in the room for the big moments.

001

Matthew Mamet @mamet.bsky.social · 29/09/2026

Rationing deletes is the odd one. One live poll is an ordinary plan limit. A cap on deleting only makes sense if someone decided old polls piling up is what pushes people to pay.

110

Matthew Mamet @mamet.bsky.social · 28/09/2026

I grew the Boston Lean Startup Circle from 30 members to 3,000 in 2010. This August the fractional meetup I organize cancelled its in-person session because three people confirmed. And I just admitted to the group that I rarely go to events. The organizer has to show up first.

010

Matthew Mamet @mamet.bsky.social · 28/09/2026

Override rate is already that number. How often the person deploying the agent reaches in and undoes what it did says more about trust than any count of API calls, and it costs nothing to log.

000

Matthew Mamet @mamet.bsky.social · 27/09/2026

Getting to a hundred first also changes which question the investor asks. At zero they judge the founder. At a hundred they ask about month-three retention, and plenty of companies look worse on that than on a story.

000

Matthew Mamet @mamet.bsky.social · 26/09/2026

Week two is when someone from sales could have caught it, and week twelve is when they first see the feature. An integration owner inside engineering still has nobody on that side to ask.

010

Matthew Mamet @mamet.bsky.social · 25/09/2026

I finally counted where TripAdvisor's product org went. 170 people, 148 companies, 21 chief product officers. For junior PMs, the best predictor was being there when two big bets went opposite ways. Tenure didn't matter. buff.ly/XyP5r9n

matthewmamet.com

TripAdvisor Wasn't a Mafia. It Was a Coaching Tree.

170 TripAdvisor product alumni, 148 companies, and a shape that looks nothing like the PayPal Mafia.

010

Matthew Mamet @mamet.bsky.social · 25/09/2026

"The model said" is already filling the slot that "a vehicle struck" filled for cars. Whoever shipped it drops out of the sentence the same way.

001

Matthew Mamet @mamet.bsky.social · 24/09/2026

Elena Verna wrote that it is a great time to be freemium. At EditMe, our free month told us almost nothing on its own. The signal that mattered was whether an account had a second person editing by day thirty. Single-editor accounts churned however generous the trial was.

010

Matthew Mamet @mamet.bsky.social · 22/09/2026

My honest split on the people talking about AI: 80 percent have read about it, 15 percent have opened a chat window and uploaded a document, 5 percent are building. I spent years in the first group. It changed at EnergySage, when I pushed a team to build automated solar quoting.

000

Matthew Mamet @mamet.bsky.social · 22/09/2026

Three percent is an average. Nobody using the feature experiences an average. If the misses cluster on the questions people act on it is a recall, and if they land on trivia it is a footnote.

000

Matthew Mamet @mamet.bsky.social · 21/09/2026

The first casualty is the careful cold email. A founder writing ten considered notes a week now lands in the same folder as a bot sending ten thousand, and nobody opens that folder.

000

Matthew Mamet @mamet.bsky.social · 20/09/2026

Which argues for a shorter list rather than a lower bar. At Embarc we doubled revenue by repositioning into one industry instead of chasing everything, and the quality problem went away with the volume.

110

Matthew Mamet @mamet.bsky.social · 20/09/2026

Half at thirty percent works until the thirty percent version is the one somebody sees and reads as the standard. I have not found a way to mark a draft as a draft that anyone actually reads.

122

Matthew Mamet @mamet.bsky.social · 19/09/2026

I do not know how much of that is the model and how much is which defaults each harness fights. I draft against a voice profile built from my own top-performing posts, and nearly every rule in it exists to cancel a register rather than to add anything.

010

Matthew Mamet @mamet.bsky.social · 18/09/2026

Alibaba ran 200M+ AI-native orders through Qwen in a single Spring Festival. Walmart's checkout inside ChatGPT converted 3x worse than Walmart's own site. Same technology. The difference is whether the agent is already standing inside the trust the buyer built somewhere else.

000

Matthew Mamet @mamet.bsky.social · 18/09/2026

I keep underestimating the consolidating half. Three tools hold three versions of the same note and none recorded which one you meant, so the merge is a judgment call before it is retrieval. I have not found a way to hand that over, local or not.

010

Matthew Mamet @mamet.bsky.social · 17/09/2026

Stripe's internal AI agent is used weekly by 10,000+ employees. Their finding: the hard part was governance, not model choice. I run the same architecture at a scale of one, and I had to build the governance layer from nothing. Neither of us tested it inside a company that's a mess.

000

Matthew Mamet @mamet.bsky.social · 17/09/2026

Picking by outcome is where I have gone wrong. I still cannot tell from the outside whether someone's lesson came from their constraints or their capital.

110

Matthew Mamet @mamet.bsky.social · 16/09/2026

What I cannot work out is how long being easier to deal with keeps paying. It works while the buyer is still choosing between people. I do not know what it is worth the first time the comparison is a tool that never gets annoying at all.

000

Matthew Mamet @mamet.bsky.social · 15/09/2026

John Cutler: the consensus you see in a company is mostly people adapting to what the powerful already believe. I've started at nine or ten companies, and the window where you can still tell the difference is about two weeks wide. Then everyone files you and the signal goes.

000

Matthew Mamet @mamet.bsky.social · 14/09/2026

I asked a guy who bootstrapped an agency to $15M and sold it whether I'd already missed the AI window. He said no. Then he talked me off my own plan: you can build a product agency with no developers, and you get smoked the moment the work turns technically serious.

110

Matthew Mamet @mamet.bsky.social · 14/09/2026

The board version is telling whoever owns the number that of course they would say that. True, usually. And also the only person in the room who read the data.

000

Matthew Mamet @mamet.bsky.social · 13/09/2026

Building the tool stopped being the expensive part. Knowing what you want it to do never got cheaper, and that is the step most people skip even on the settings screen. The ones who will build their own are already the ones hunting for workarounds.

022

Matthew Mamet @mamet.bsky.social · 11/09/2026

I once walked into the CEO's office to recommend we shut down the product I'd spent two years building. It won an award. It was also the wrong thing to keep funding. The hardest product calls aren't about the product. They're about separating its quality from your pride in having built it.

000

Matthew Mamet @mamet.bsky.social · 10/09/2026

A founder I worked with had one constraint that reorganized her whole growth plan: it had to run inside a nine-to-three workday, kids in school. We cut everything that scaled with her hours and kept only what compounded without her. The time limit made the strategy better.

000

Matthew Mamet @mamet.bsky.social · 09/09/2026

John Cutler named the recontextualization tax: every handoff between a person, a tool, or a model costs work to rebuild lost context. I run my practice on AI agents and pay it daily. The fix wasn't a smarter model. It was making the chat disposable and the file the memory.

010

Matthew Mamet @mamet.bsky.social · 09/09/2026

Trust calibration does not carry across a release. It took months to learn where the current one fails, and a swap resets that, so a model that benchmarks better can still be worse the week you adopt it. The switching cost is confidence, and nobody prices it.

000

Matthew Mamet @mamet.bsky.social · 08/09/2026

A liquidity problem and a demand problem look identical on the dashboard: too few transactions. Add traffic and watch. If conversions stay flat, it was never demand. How I diagnose it: matthewmamet.com/blog/how-i-u...

120

Matthew Mamet @mamet.bsky.social · 07/09/2026

A founder was interviewing me for a fractional role and I talked myself out of it. He asked where the category goes once AI eats the workflow. I said I don't fully know. I'm good at scaling chaos, not seeing around corners. Better to name that than discover it in month three.

001

Matthew Mamet @mamet.bsky.social · 07/09/2026

People are reacting to the length more than to the machine. Something that took a minute to write and ten minutes to read is asking me for a favor I never agreed to. Blocking is just cheaper than reading it.

000

Matthew Mamet @mamet.bsky.social · 04/09/2026

In 2010 I hand-wrote SQL against my own production database at 2am to find out if my bootstrapped wiki was real. The number that predicted survival was 19 customers who would recommend it against 2 who would not. Rarely the one already on the dashboard.

110

Matthew Mamet @mamet.bsky.social · 03/09/2026

Etsy cut 12% and went out of its way to say it wasn't about AI. I believe them, and that's the tell. When layoffs "aren't about AI," the story underneath is usually slowing demand in the core. A marketplace is only as big as the smaller side wants it to be.

000

Matthew Mamet @mamet.bsky.social · 03/09/2026

The all-nighter is the cheaper line on someone's budget, which is the whole problem. Zero-downtime work competes with roadmap features for the same engineers, and 'we ran a 2am window' never shows up as a cost the way a slipped feature does. The invisible cost loses to the visible one.

000

Matthew Mamet @mamet.bsky.social · 02/09/2026

A studio runs a portfolio and spins the winners out. Mine is the opposite shape, one venture I keep, funded by the practice and capped so it never needs a team or my full attention. The cap is the whole design.

110

Matthew Mamet @mamet.bsky.social · 31/08/2026

CPO tenure averages 2.6 years, shorter than CEO or CTO. It isn't that the people are bad. The job is three roles stacked under one title, and it gets swapped for a different one every time the company doubles. You're hired for today and judged eighteen months later.

120