London
Joined March 2021
- Tweets215
- Following660
- Followers286
- Likes1.4K
a truly excellent and timely read.
colossus.com/article/sane-as…
@zebriez , nice job.
I'll admit, I've felt an immense degree of pressure working on what you might describe as the fringes of AI's core while at @linear, which is not a lab, but very much in the space, and therefore susceptible to the rigmarole of constant change.
Many friends are at labs, and they've clearly all adopted a similar kind of psychology around money, output, intensity. They all possess this "it's now or never" kind of mindset. Maybe they're right? Maybe that's what you need in the AI arena. But also, maybe you can work with extreme intensity without it becoming you.
I'd been feeling a lot of pressure to adopt the same mindset. This piece from Brie landed at the perfect time. So, thank you.
Matthew Roberts retweeted
I see such a wide range of AI adoption curves in engineering orgs, it's fascinating. This could be the widest gap we'll ever have, which also makes sense given how big of a shift we're seeing.
Right now, I think companies are in one of five stages. And you'll have to take off your X glasses, because all of us yapping on here aren't quite living in reality.
Which btw, I like. Many of us are the crazies living in a future that could be a decade or more away, and that's fun to do, together.
That being said, here's the stages I see engineering orgs in today. And before you read this know, this is just my (loosely held) opinion.
And as I've said time and time again on here, but feel I always need to repeat. I'm not an expert, I'm just an insanely curious guy who loves sharing what I think.
So on that note, here's my thoughts on the five stages eng orgs are in right now:
Stage 1: Humans still writing code like it's the stone age.
- there are less and less companies in this stage by the week, people are moving out of this stage quickly. But yes, believe it or not, some companies still only trust humans to write the code.
Stage 2: One lab, one model, one effort level, and usually JIRA for ticketing.
- this is the stage where an engineering leader has either decided their team needs to use AI, or an exec told an engineering leader they need to. They usually just go straight for Claude Code, and everyone on the team uses Opus High. Tickets are managed in JIRA, just like they always have been, and humans manage those tickets.
Stage 3: Multi-model, driven by a third party model router that picks the model and effort level. Actively building a software factory.
- usually people in Stage 2 graduate to this. They start getting crazy high bills from their model provider, and realize they have no idea how to tell their team which model or effort level they need to use for their work. They get pitched the concept of saving 30% or 40% if they switch to a company that offers a model router and decides which model and effort level to use per task. And, in this process, they start building a software factory.
Stage 4: Multi-model, driven by internal benchmarks based on real PRs and their actual codebase. Software factory is mostly built.
- usually tried auto model routing, but realized it wasn't really optimized for the work each team does. Started doing internal benchmarking and realized that the model and effort level the data science team needs for their python coding, is different from what the data engineering team needs for their python coding. Comes up with model/effort combos by team, but has a hard time keeping up with it. Software factory is mostly built, either moved from JIRA to Linear and now actively hands off tasks to Linear, or moved off a formal ticketing system entirely.
Stage 5: feels the most behind, constantly optimizing models and effort levels, most easy/medium tasks are done entirely end-to-end by agents. Fully functioning software factory, but constantly updating and trying to adjust to new model releases.
- yes, the first few words here are important. The most advanced stage feels the most behind. Kinda like Einstein with Math. He was quoted as saying, "the more I learn about math, the more I realize I don't know about math." The engineering orgs that are the most ahead, often feel behind. Terra 6.1 comes out, they're scrambling to figure out what they should use it for, where it will save them money/time. Some teams love this excitement, want to constantly change, for others its dizzying.
Okay, Wednesday morning brain dump complete, time to play some WOW Forever. Zug Zug.
Matthew Roberts retweeted
Somebody should say a few words.
This profession has been lamenting in fragments for two years with a headstone that turns out to have the wrong name on it.
@davekiss delivered a masterclass at Resend Forward. Listen, grieve, and leave renewed with hope.
Matthew Roberts retweeted
Often simplicity just works.
Companies moving over and now creating their docs in @linear
Watch the s-curve on this kinda of stuff kick off. It'll work for a bit, people will all start doing it, and then it'll become an anti-signal.
When a cake with my dogs name on it is handed to me by someone with my mother's maiden name while I'm walking across hyde park on a Sunday morning, just know, I will never use your product.
This stuff will only continue to work if you have proof of human work behind it. Film yourself doing the research, discovering the delightful moments, and really engineering that bespoke human interactions.
Send that to the person you're trying to reach, the effort will signal more than the output from it.
It's like all those personal letter companies popping up in the early startup era. The value was never in the letter itself, it's that someone important took the time out of their day, used their own pen, and put their own thoughts into something.
F your AI b2b SaaS.
We just launched Palantir for booking meetings.
The last 14 days, companies like @MonacoGTM , @Nowadaysai, @Rippling , @salesforce , @marketercom, @posthog , and @retellai spent $100k+ with us on literal cakes.
From today, Daymaker finds your prospects, reads their entire digital footprint, buys the one gift that makes their day, and gets you the meeting.
Comment your name and I'll reply with everything Daymaker finds on you and the gift it would send you. If you dare. 🎁
This has always been true of startups, and it's true of plenty of the wider economy too. Carmakers lend you the money to buy their cars, and banks finance the homes whose prices depend on bank lending. So what?
The only real risk is a loop that never delivers value to end customers, so revenue never takes over from VC money as the thing keeping it going. Cloud and mobile both started as VC-funded loops and became self-sustaining.
There's almost certainly some form of adopter bubble in AI, and it'll take some time for the technology to diffuse outside of that group, and there will be losses and failures in that process, but that's the nature of VC funding. What do you expect when most VCs have to return funds to investors in 10+2 years and not 100 years? The incentive of speed is what causes both incredible outcomes and failures.
There's nothing new here.
So let me get this straight.
- VC funds Company A.
- Company A uses that capital to buy AI services from Company B.
- Company B uses that revenue to justify a higher ARR multiple and raise more capital.
- Company B spends its new funding on AI services from Company C.
Rinse and repeat.
If the estimates are accurate, roughly 70–75% of AI spending is coming from unprofitable, VC-backed companies.
How much of the AI industry’s explosive revenue growth is ultimately just venture capital circulating between companies?
And more importantly, how much of today’s AI valuations are being supported by revenue that disappears when the funding cycle slows?
“AI is an accelerant, it makes more of whatever it is. If you do not first identify the critical few, AI simply multiplies the useless.”
True for most use-cases, but particularly true for product development. Adding an extra feature doesn't inherently make your product better.
Matthew Roberts retweeted
And now you can preorder the NOMAD LETTERS (out December 1!): press.stripe.com/nomad-lette…
For years these letters have circulated (at first, unofficially) among investors, acquiring a devoted readership. We wanted to give them a permanent home in print.
Nick and his partner Qais Zakaria’s record is extraordinary. The letters are the account of how they got there, featuring everything from studying Amazon and Costco (Nomad’s defining investments)—to visiting cement plants in Zimbabwe, and beyond.
They’re candid about their mistakes and curious about the forces that shape investors’ behavior: incentives, psychology, and the wisdom (or foolishness!) of the crowd.
Since closing the fund, they’ve devoted their time to philanthropy, a journey Nick recounts in the lovely foreword he wrote for this edition.
I’m thrilled to continue adding to the investing corner of our catalogue, and I can’t think of a better work to follow Poor Charlie’s Almanack. ❣️
Matthew Roberts retweeted
Linear is coming to Europe, and we’re starting in London.
Join us on November 10 to celebrate the launch of our new hub and spend an evening with the engineering and product leaders building some of Europe’s most loved products.
luma.com/linear-london
I'm not a fan of this new outbound tactic of sending direct invites on gcal. Please make it stop.
Relatedly, I don't know whether it's because of AI, better enrichment tools, or people have just gotten more capable at targeted outbound (probably all the above), but my email inbox at Linear is virtually unusable. 623 unread emails, 90% of which came in over the weekend, and almost all of them are cold outbound.
Classic channel s-curve in action. It works, people rush to it, saturate it, and then it no longer works.
Having grok bot tell me what to work on isn't helping either, because I just end up ignoring those pings also.
Good marketing has always been about amplifying the characteristics of a really good product.
We often forget that and try to reinvent marketing from first principles every couple of years.
Hearing more and more people in tech simply falling out of love with the space.
Maybe we’re just coming to the end of the romantic era in technology, where scarcity (of craft, skill, insight) is what enabled those beautiful eureka moments.
This happens in almost every category given enough time, where the age of heroes comes to an end, and some form of industrial era begins.
I think this is ultimately a good thing for the whole. Scale and standardisation will lead to more useful things being produced for the world.
But it’s clear we’re losing that sense of the exceptional individual, and people are leaving tech to seek it, or they’re actively picking up more hobbies and finding it elsewhere in their life.
Matthew Roberts retweeted
Ticket requests for Orbit (by Linear) are now open.
Join us in San Francisco on Dec 9 for an exploration of the future of product development: If we can produce more software than ever, what does it take to make it good?
orbit.linear.app
Matthew Roberts retweeted
As you invite more agents to our work, and create software factories to be more productive than ever, you’ll soon outsource everything to an agent, even the thinking.
The work you will do will feel frictionless, it will feel amazing on the surface, but once you go a bit deeper, it’ll feel extremely shallow. It won’t reflect the struggles you went through, your decisions, and ideas.
It will no longer remind you of you and the work you aspire to.
You will turn into just another extremely efficient producer of things, but not a creator like you were before.
Invite friction to your work.
Matthew Roberts retweeted
Actually the chronically negative you can identify in hiring and avoid
The real damage to startups is done by 'Smart Skeptics'
Who are smart, sound even smarter than they are, are great at raising objections, can hold great intellectual conversations, have a lot of anecdotal evidence about everything.
But can't execute/move metrics/ solve hard problems even if their life depended on it, because they are skeptical about everything.
And they dissuade the believers as well.
Avoid them like a plague
Matthew Roberts retweeted
Whoever you paid for the fake engagement, you should get a refund
The comments are too blatant, too lazy, no finesse. They’re like bad plastic surgery. “Those can’t be natural”
I’m sure they sold you guaranteed virality and charged a boatload. But being associated with fake engagement is not worth the bigger numbers
Introducing SubQ Snap, the first long-context decision model for real enterprise work.
Snap is the only decision model or LLM where compute per token stays constant as context grows.
• 384K tokens of multimodal context, 12x larger than Jev.
• 61.7 on Decision Index v0.2.1 vs Jev’s 57.9, outperforming on 13 of 16 enterprise benchmarks.
•Processes 100K tokens and returns a decision in under one second.
•$0.05/MM input tokens. Output is free.
Available today to anyone.
The first 10,000 sign-ups get 100M free input tokens.
We’ll likely hit capacity fast and will add users as we add compute. Compute offers welcome 🙂.
Latency will be variable as we tune the platform to this new model and category of users.
Model speeds are 80ms for minimal context, 250ms for 32k, 660ms for 100k, and 2.5 seconds for 384k.
Visit: subq.ai/
Matthew Roberts retweeted
Saying software is dead because anyone can build it is like saying photography is dead now that everyone in the world has a camera in their pocket
Not only is it not dead it's the beginning of the Kodak Brownie moment for software, and we will see an absolute explosion of it
It will be the first generation for who building software will feel no different than using it. This is crazy!
The quality and bar of what we expect from our everyday apps will be raised dramatically, we'll have personal software, we'll have disposable software, and agents will be the foundation for most of it
Tech has always been chaotic and evolving, the people who have been able to adapt and move in front of new opportunities have always excelled, this is nothing new, I'm seeing it now and we'll see it 10 years from now
Find the opportunities in the chaose and disruption, there are countless of them
Couldn't have said it better myself.
The Bull Case for Linear Being a $100BN Company:
"Linear is the clear winner.
They have built an agentic product first that allows us to build 100x more software, and that means 100x more features than ever before.
Humans cannot keep up with it, and humans still have to work with agents. If every human on your team's gonna build 500 features and 1,000 issues and you have 10 people on your team, you need a process and a system of record for managing all these issues with your agents.
We need a new system of record for it.
The team at Linear has figured it out. We have seen an explosion, 50x more agent usage than 90 days ago.
This will seem cheap when Replit's at $15BN, Lovable's at $100BN, because Linear will be the one powering them all". @jasonlk
Love to hear your thoughts, what is your bull case for Linear @karrisaarinen @stephzhan @adambain @dickc @soleio @zoink @Mkclements