Sandhya

@trq212 Would not taking up all available memory be one of those details

Agent view is the best Claude Code native way to manage multiple sessions, kind of like tmux built for CC. We spent a lot of time getting the details right, I hope you enjoy it.

Fabulous post on the PM craft, couldn’t agree more. Especially the fact that AI as a source of leverage has shown us how mediocre most PM work has become. I see so many who have completely lost touch with their product’s actual full user experience, from facing friction in the moment of consideration to yelling at AI bots that don’t work.

@AndrewCurran_ Hm. So the PE firms are going to sell their own AI service to the companies they have a stake in..? Yet another value extraction channel I guess. Nothing could go wrong for the employees.

Official announcement from Anthropic:

Just realized that the last TBPN post to pop up in my feed was their acquisition, they might it as well have retired if you ask my feed

building AI evals right now feels like CI/CD in 2012.. everyone knows they need it. everyone agrees on what great can be in theory but almost no one actually does it... and the tooling is just barely past the vibes stage. the companies that figure out eval pipelines (converting production issues into real tests) will ship better AI products 10x faster than the ones running on optimism. it also take a certain mindset to do this well... most teams just don't want to admit their baby could look prettier .. their evals test for life rather than excellence

@avemii It's a leakage bug not supposed to happen

@sandhya I dunno how much memory this Claude beast needs

chat am i the only one (and no this wasn't a computer use session) getting tired of the unending bugs

@jpschroeder Summoning @simonw to this 🤯

what. what. what. gpt-image-2 almost passes the pelican test...in a screenshot of a code editor.

When you have the kind of PMF Anthropic has, the “growth” team’s job is to not f*ck it up. That’s it.

Yes I am selling an addictive drug but will NOT give you a taste sir

Easily the most ill considered AB test I have ever experienced - removing Claude Code from the Pro subscription for new users!!!!

Increasingly hearing (and feeling) the word “addiction” being used by ai-pilled developers when describing our daily work.. mistaking output for progress and impact the way you mistake euphoria for happiness. I want my moments of friction to function as thinking breaks - forcing me to question whether something should exist rather than shipping. Building /= Shipping. The latter is a means to the former end. H/t @mitsuhiko

Friction is also how you learn and build judgement! Loved this talk. I work with teams on improving quality with evals and monitoring and we always see engineers eager to automate the process before they understand it. You can’t express taste and judgement without first putting in the manual work and creating the ground truth that cloud agents can use to automate your systems. No pain, no gain.

This is the case for demoting yourself. All my client CEOs seem incredibly frustrated with their managers right now. As a manager/leader, one of the most important roles you have is to define a clear picture of what success looks like for your team. This is hard to do when what's considered good in those roles has fundamentally changed and you have never seen or done this new job yourself. Ever. There are two options: going back to the builder path or blowing up your manager path. If you are managing a relatively small team and always considered yourself more of a builder than a manager - embrace the super IC path and reinvent yourself. Even when I had a team of 60+ people at a startup, I considered myself mostly a great IC, a mediocre manager and tried to lead by example. If you are like that, embrace your craft. The career opportunities for AI builders with judgement and security awareness are boundless (and lucrative). If you are not really a builder, consider the fact that the skills that served you well in the past (managing up, getting headcount, navigating politics and big egos) are not going to quite be as valuable in tech going forward. Stop asking for headcount and disrupt your own team. Have 15 direct reports instead of 5. If you were the VP before, you are now also the Directors. Let go of ICs who are not reinventing themselves. Hire fresh AI-native talent. Get closer to the technical core of your company - which currently sees you as an obstacle to overcome instead of support for what their are building. It might sound counter intuitive but demoting yourself might be the best thing you can do right now. Consider doing it before it is done to you by the CFO's new budget.

"The ones who were the best at working in the past, the ones who mastered the old game, will find it the hardest to go through this reinvention stage." Every PM needs to listen to today's episode with @nikhyl. This is the most honest, real-talk conversation I've had about what's actually happening to PM careers right now. We discuss: 🔸 Why the next 2 years will be the most chaotic in PM history 🔸 Why half of today's PMs won't survive the shift 🔸 Why the fancy logos on your resume now matter less than ever 🔸 The 'smiling exhaustion' he's seeing across top PMs 🔸 His prediction: companies shed 30,000, rehire 8,000 — all AI-first Full episode here on X 👇 Also find it on: • YouTube: youtu.be/yUohoaC8_Hs • Spotify: open.spotify.com/episode/6OO2GJ… • Apple: podcasts.apple.com/us/podcast/why… Thank you to our sponsors for supporting the podcast: 🏆 @WorkOS — Modern identity platform for B2B SaaS, free up to 1 million MAUs: workos.com/lenny 🏆 @TrustVanta — automate compliance, manage risk, and accelerate trust with AI: .

I upgraded my Claude token counter tool to compare different models and Opus 4.7 does appear to use 1.46x times the tokens for text and up to 3x the tokens for images - it's priced the same as Opus 4.6 on a per-token basis so this is actually a pretty big price bump

The irony. Just as creating web apps became trivial, user preferernce switched to text files for their agents. That probably covers 80% of apps getting vibecoded.

Copy paste /go

@tanaypadhi Atoms updating their deck to focus on nuclear energy

how did Allbirds pivot to AI compute hardware before the shoe company literally called ASICS

strong agree.. i think we started down this path as soon as the ai labs started post training models specifically for coding and long horizon task completion. Layer on a more efficient custom harness over it and the cost of rolling your own suddenly feels like a business decision that needs to be justified - is it going to be more performant? cheaper in total token consumption? If not - what's the point?