The weekly top 10 for B2B tech operators · Every Friday
Done - check your inbox to confirm.
Revenue & Burn per Employee. In this profit-over-growth new SaaS world, forget vanity metrics. This post makes the case for tracking Revenue and Burn per Employee as core performance metrics. Why? They reveal whether you're scaling efficiently- or just adding bodies. It also includes great benchmarks across stages.
Reverse Trials are a play on freemium, where new users start with a time-limited trial of all your paid features, and at the end of the trial, they can either buy or downgrade to a fully free tier - this article also explains how Airtable does this well. The benefit here is that, emotionally, the user experiences loss aversion, where the pain of losing something is twice as powerful a motivator as the pleasure of gaining.
Not sure if you were impacted (I certainly was), but the PHPShell issues last week impacted some 70+% of the internet, but it wasn't just about the vulnerability itself; it was the speed at which the hacks are coming in- they are coming in thick and fast.
Amazon now runs over 1 million robots alongside 1.5 million humans. The robots took the heaviest, riskiest jobs, and severe injuries fell 40% in its robotic warehouses. But strain injuries from repetitive work rose 77%, and the professor who co-wrote that study says it's because the robot now sets the pace.
New term for you to ponder, "Leading from the front" isn't just a military cliché for most of us in startup land, though - it's the difference between high-trust teams and checked-out ones. Stay SaaSy breaks down how great leaders model urgency.
Most business plans fail due to a skill set of founders: over-optimism, as well as a lack of adaptability, and inadequate market understanding. Founders often overestimate their control and underestimate challenges, leading to misguided strategies. Recognizing these pitfalls and fostering a culture of flexibility and realistic planning are the big skills for success - own the hard stuff, and never ask for what they won't do themselves.
New one for our AI workers - Attention. Ever wondered how a model knows what 'it' refers to in a sentence? Every word gets checked against every other word to decide which ones matter. Natalie Wolchover's Quanta column explains queries, keys, and values in plain English (no maths degree needed).
Adding onto #6 above - Remember Meta's code changes up 220% and incidents up 40% (see #4 here). More code, more outages. AI agents have roughly doubled PR frequency at almost every tech company in a year, data shows out-of-hours pages nearly doubling too. So now there's a whole category of AI SREs promising to investigate the incident before anyone wakes up. SRE's read the alerts, logs, and dashboards and hand you a likely root cause in Slack. Stripe's in-house one averages 3 minutes for a quick investigation, 14 for a biggy.
Fast follow from #8 above - if agents are writing twice the code, who reviews it all? Anthropic's Claude Code team (Claude writes \~90% of it) sizes human review to risk: docs and flagged changes need no human, normal work gets one, the riskiest paths get two. And a bot's sign-off never counts as a person's. Here are their six layers of checks.
Following up from last week's Bending Spoons case study (see #10) - a different kind of holding company. Long Lake has raised $7.5B to buy services businesses (residential services, architecture, HR, and now Amex GBT in a $6.3B take-private) and put AI inside them. Their pitch: selling AI from the outside means fighting security reviews, adoption, and pricing, so own the business instead.
Get the top 10, every Friday
Curated SaaS and tech insight from around the web, repackaged for people to put to good use — free.
Done - check your inbox to confirm.
WIN RATE: This is calculated via closed-won deals divided by qualified opportunities. It's easy to track, and also easy to ignore while everyone stares at their pipeline. According to Jason Lemkin, Owner.com took theirs from 38% to 53% in 10 weeks by having AI score every sales call. Check this list of ways to move it runs from pricing and qualification to the deals that ghosted you.
One for the Digital Marketers - Your iPhone retargeting numbers might be about to look weird. AdExchanger reports iOS 27 is quietly blocking hundreds of programmatic data companies across every browser on the phone. Is Apple now a shadow ad regulator?
Corporates are now also on the Cap Table for almost every big AI round. About 88% of US AI deals have some kind of corporate (BUT - Nvidia wrote most of that). Looking at KPMG's Q3 Venture Pulse - corporate VC hit $532.8B at the end of Q3 - already a record year, and 14 deals above $1.5B. PitchBook's new note also prices market entry into the AI frontier at about $100B over five years.
85% of enterprises are working on agentic AI, but only 5% have one in production that pays its way. Uber's agentic coding usage went from 32% to 84% across \~5,000 engineers in a month. The year's token budget was gone by April. That's one of the cases in Ray Rike and Peter Buchanan's rundown of the six things stopping agentic AI from paying off.
This one is for you early-stagers - Sylvia Huang's guide to early-stage viability lays out what to watch at each step. Good stuff in here - from pre-revenue intent through retention, repeatability, and unit economics to tell them apart.
Tunguz's 2026 survey of 455 senior technical builders. Cloudflare hosts 62% of agent sandboxes. 45% route across models. 48% review AI logs with coding agents. And security is now the #1 headache, up from sixth last year.
If you're selling agents by the seat, your buyers have moved on. G2's 2026 survey of 1,019 B2B decision-makers - 57% now prefer pay-as-you-go (38% last year), 19% pay-per-outcome, and only 9% call per-seat the fairest model. And 30% would consider switching vendors for better agent capabilities.
Your next round may get crowded by investors you already have. Hunter Walk expects more pro rata conflicts - because early rounds are bigger, rounds land faster, Series B and C can still return 20-100x in the best cases, and have buyers ready to go in Secondary Markets.
The default fix for weak sales is more features. So check out Sylvia Huang's checklist, which tells you whether that's right: prospects who vanish after asking about price or stall-out while onboarding point to the sales process. Users who buy, skip the core workflow, and never renew point to the product.
Bending Spoons is basically a holding company for once-great tech that's been put out to pasture: Evernote, WeTransfer, Vimeo, Eventbrite, AOL, and now Airtable and Miro too (both bought for around 90% below their peak valuations). Lemkin pulls apart how they run them - about 60 people on Vimeo, and AI writing 90%+ of the code. All part of a weirdly profitable portfolio: Q2 revenue $704M, up 126%.
I bet if you asked any founder their growth rate, you would probably get it instantly. But if you ask how long it takes to earn back what they spent winning a customer.........crickets. But that's the number investors check first. Median B2B SaaS is 16 months, top quartile under 6.
Ever feel weird running a startup like it's a friggin' classroom? If you nodded your head, you're not alone. This piece unpacks "educational trauma" — why many leaders default to teacher mode, and how to build an adult-to-adult culture instead.
Anyone who's raised here has heard the dreaded phrases: "Too early," "too crowded," "not in our thesis" - VC-isms. This article breaks down the _real_ reasons VCs say no. 10 rejection themes, decoded with some brutal clarity. A must-bookmark/read in prep for your next raise.
Code review is DEAD! Here are 16 things Thorsten Ball (he literally wrote the book on building interpreters) believes about where software development is heading. A few that'll start a fight in your engineering channel: code review dies, unit tests might too, and most bugs won't be coding bugs - they'll come from asking for the wrong thing.
Think $250K ARR per head is great? According to this ICONIQ report, the top AI companies (at $100M+ ARR) do $655K, top quartile is $890K. That's a $200M company running with about 300 people vs. 800. Also, at $100M+ ARR, the median grows 115%, keeping 90% of revenue before expansion, as it does at $655K ARR per employee.
Tech word of the year probably - Tomasz Tunguz argues that AI inference is about to become the most important market in software, and that every app is about to become an AI reseller. Inference will hit about $350B by 2027, nearly 2x the database market. If AI usage starts eating half your revenue, what's left of your margin?. The main models are commoditizing though, prices are down 41%, and token use up 50% since July.
Fast follow from #6, here's that margin squeeze from the inside. If you're pricing an AI product right now? Read this - the real unit economics for the AI product Candor: $2.50 to run an interview, $7 to buy one, and a free trial that costs him $68 a signup. The whole pricing model is in a public spreadsheet (you can copy). Oh but here is the gotcha nobody budgets for - upgrading the model made the product better but burned 25-35% more tokens.
a16z's Top 100 Consumer AI Apps is out, and this time it adds actual US card-spend data - which is key! Only 4.5% of US consumers pay for ChatGPT, Gemini, or Claude. The top 1% of payers average \~$903/month and outspend the entire bottom half combined.
Another follow from #6 - if the models are commoditizing, what's left to defend? Dan Hockenmaier has a one-hour Moat test for ya! Find the compounding advantage - switching costs, network effects, economies of scale, or data. Speed, brand, a great team, and funding don't count (he calls those "fuel"). Then check what the team actually works on every day (not what the strategy deck says).
YOU SUCK - 12 companies whose first product flopped. YouTube was a video dating site, so few users that it turned to Craigslist. Slack was the internal chat tool inside a failed game called Glitch. Bill Kerr's roundup of first products that were dumpster fires.
CLTV to CAC Ratio: This is the classic signal of go-to-market efficiency. Everyone quotes 3:1 (too low, likely not PMF; too high, you ain't spending enough on marketing) — but this breakdown goes deeper with benchmarks by sales motion (PLG vs. enterprise), contract size, and stage. Bonus: includes a free model template to run your own numbers.
Testing new marketing growth tactics takes a lot of resources, and most of us often don't have much time to run experiments. So check this Google Doc from Dashly where they collected 100 growth marketing hypotheses tested by a bunch of their experts. (includes ad retargeting, wait list for product launches, niche glossaries, etc).
Don't be surprised if you land a sales rep from a big company who met quota at their last job but gets 0% with you - these kinda reps succeed on product-market fit, tight process, and real support. Take away any one of the three, and they close nada, because they'd have to invent the sales process deal by deal. The fix is in the article - you gotta click it!
John at SuccessVP hosted a dinner of CROs and CCOs in New York in September. They're still struggling to find the AI productivity gains. His read: the models aren't the bottleneck anymore; deployment is. Someone has to connect the context, build the evals, and define what done means.
The competition continues, and a16z's David George thinks the AI model rankings no longer matter. We can swap models task by task, so switching costs are gone and pricing is a race to the bottom. What's left is who creates new customer behavior and who owns distribution. (His bet is OpenAI on both behavior and distribution - I'm not sold on that).
A lot of comments on the meat puppet post last week. So follow up for all you Meat Proxies - do you still enjoy it? And on the positive side - good thinking here on saying goodbye to the hard parts that never really mattered with your job anyway.
Follow-on from # 6 above. As a former Systems-and-Design-Thinker-for-a-job (and Ecologist), I'll take any excuse to quote Donella Meadows. Tunguz gives me one: Donella says writing code by hand is no longer economically productive, so what's left is the system. Also - let's take the three properties of great system design, resilience, self-organization, and hierarchy, and apply them to AI coding. The code writes itself now. The job is designing the loop that decides whether it's right.
Cut a dollar of brand spend now, and BCG reckons it costs $1.92 to win back the lost market share later. Google calls brand cuts an expensive loan against future pipeline. 67% of CMOs trust their marketing mix model to predict sales. Only 40% of CFOs do, in a BCG survey of 81 execs. Google argues this gap is why brand budgets get cut first.
Metrics don't get you funded. Harris (ex-YC partner) has seen nine-figure rounds close at $200k in revenue and companies fail to raise at $10M ARR. Be strategically persistent - love the take that "prepared minds get built in casual conversations".
This article tracks six liquidity events, from first check to last share, then runs IPO against M&A.Snowflake's lead VC took 9.5 years from first check to seeing a return, and that's the best case. Rubrik's lead is 12.4 years in and 82% out. Wiz went from founding to $32B in cash in six years.
Get the top 10, every Friday
Curated SaaS and tech insight from around the web, repackaged for people to put to good use — free.