Top 10 in Tech — A weekly top 10 for B2B tech operators: SaaS metrics, pricing, retention, GTM, fundraising and product strategy. Published every Friday.

The weekly top 10 for B2B tech operators · Every Friday

376 issues · Every Friday since 2018 · 09:00 NZT
Issue 376

Permalink →

SaaS METRIC OF THE WEEK

ARR alone doesn't tell you what kind of growth you're getting. Check this article (and dashboard) breaks MRR into New, Expansion, Contraction, Churned, Net New, and Ending. A company growing through expansion behaves very differently from one that depends entirely on new logos. Only the Excel bit is paywalled. The diagnostic logic: if trials are stable but activation drops, fix onboarding. If activation is good but the paid conversion flat, the value moment is unclear. If the active base grows faster than acquisition, retention is doing the work.

JOBS 1/2

Pragmatic Engineer has a two-part State of the Software Engineering Job Market 2026 ( here for part 2). It's way less doom and gloom than we assume (see below for more). Meta grew software engineering 20% over two years, then laid off 10% last week. Apple up 10%, Google up 5%, Microsoft and Amazon flat. But Scaleups grew faster: Ramp +94%, Wiz +84%, Datadog +68%, Rippling +55%. AI engineering roles are way up - 50-100% YoY at large tech firms.

JOBS 2/2

Benedict Evans following up on the doom and gloom bit above on why AI job-exposure analyses don't work. The test case: 50 years of financial automation (VisiCalc to ERP to cloud). If you'd done a "professions exposed to automation" chart anytime since the 70s, Accountants would have been at the top. Instead, the accountant headcount kept going up. Just look at the Big 4.

CASH

And shit loads of it - how are we all not screwed? Google generates $174B in annual cash flow - literally the best business model on earth - and it's not enough for that hungry hungry AI. They're spending more this year on AI infrastructure than the world's best cash machine can produce - their solution - an $80B equity raise (the largest ever), $100B+ in total debt up from $25B a year ago, a 100-year bond at 6.125% (matures 2126) - WTF. Anthropic and SpaceX (now an AI company) and OpenAI are all listing REAL soon - so much cash about to be thrown around.

QUOTA

Interesting idea - Your VP Sales (hopefully) has a quota. Your VP Marketing probably needs one too. Check Jason Lemkin's classic 3-question hiring filter: 1) What was your lead/opportunity commit at your last company, and how was it determined? 2) I want to hit $Xm ARR by year-end - what do we do in the first 90 days? 3) How should sales and marketing work together to hit it? If they fumble on this, they prob never owned a number.

GROWTH

Another Lemkin's post: As discussed in prior weeks' posts (see #9 on Productivity), AI broke the link between revenue and headcount. When Google crossed $30B in revenue, it had 32,000 people. When Salesforce did, it had 79,000. Anthropic crossed it with 5,000. And the Time-to-$1T is even more crazy: Apple 42 years, Google 21 years, OpenAI \~10 years, Anthropic \~5 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.

Issue 375

Permalink →

SEED

Some more analysis on seed stage deals this week. Pavel Prata looked at the 10 biggest VC funds ($10B+ AUM) across three VC/Tech industry "periods": the pre-pandemic SaaS years (2015-19), the cheap-money boom (2020-22), and the AI era (2023-26). Collective early-stage deals went from 140-150/yr to 370-400/yr. a16z: 16.6 to 48.7 to 75.3 deals/yr. General Catalyst: 15.2 to 31.7 to 61.5. Both absolutely sending it!

ONBOARDING

Railway gets about 10,000+ signups a day! To manage this volume, they killed their generic welcome sequence and went event-triggered. Old open rate: 27%. New: 50-70% in the first 24 hours. The strongest signal: trust center document downloads (SOC 2 reports) - got them about a 50% reply rate when sales reaches out.

BURNER

Here's a cool concept for our tech dictionaries - work-streams that need heat! In the cold-start problem world of early-stage traction, we need the Venn diagram overlaps among build, distribution, and founder capacity. Founders normally sequence them, but the argument is that they should run in parallel. Distribution is the one left cold first - it can run weeks with nothing visible to show. By the time you need it, you've lost months of compounding momentum.

TOKENMANIA

The flipside of the agentic coding surge is the 2026 bill shocks. The WSJ reports Uber already burned through its entire 2026 AI budget (set in late 2025, before any of this actually worked). Amazon pulled an internal leaderboard that was driving too much usage. Salesforce went the other way, and 10x'd its budget. Uber's COO on a podcast: spend is high, specific gains hard to point to. Yeah, we all collectively coulda told you that. But apparently, 94% will keep spending on AI even when it fails.

AI DOCTORS

Exploding Topics surveyed 1,065 Americans on AI health use. 66% have tried AI for a health query. 25% report a "serious problem" after following AI health advice, another 20% report a minor problem. Among those who used AI for mental health or stress, 41.56% reported a serious problem. 82% feel "listened to" by AI vs 74% by human doctors.

AI SKEPTICS

Tomasz Tunguz maps out where the market is actually shorting AI. Median short interest across software, semis, data centers, and hyperscalers is up ~24% in the last quarter - because we all know that shit is expensive - as it's concentrated, these shorts are a bet against companies whose AI exposure still depends on future capital, future demand, or future operating leverage. AI cloud/neoclouds: 16.8% of float, which is up 60% in a year. Dev tools/infra software: 9.5%. Enterprise SaaS/AI apps: 8.9%. Hyperscalers: 1.1%. NVIDIA: 1.2%.

BACKPRESSURE

Another new entry for our Tech Dictionaries - it's a systems-engineering term for downstream refusing more work until the producer cleans their shit up. Lucas Costa argues it's the next frontier for AI-aided dev work. Right now, the human is the backpressure: reviewing, prompting, and copy-pasting feedback between the coding agent and the review bot. The fix is tests, types, benchmarks, review agents, and PR monitors that send bad patches back before they become a human's problem.

POSITIONING

Been through this a bit recently - April Dunford's rule for company/product positioning in the AI era: ignore future competition in your positioning, but NOT in your roadmap. Position against the competitors actually showing up on your prospects' short lists today, not the fast-moving threats you're convinced will arrive. But also build the product that can win the future match-up. Hype the vision for markets and investors. Sell against today's reality.

CASE STUDY

AMAZON - Bloomberg took a deep dive on Andy Jassy (CEO at Amazon): since ChatGPT, Microsoft, and Alphabet have collectively booked $600B more in future business than Amazon. Jassy's response: $50B committed to OpenAI, $13B+$20B option to Anthropic (with Claude now live), $200B AI infrastructure spend this year (they are becoming the Amazon of AI), $25B Mississippi data center cluster, 60,000 layoffs, dozens of projects killed.

Issue 374

Permalink →

BUBBLE

Let's check in on this article in a year - see how it aged. OpenAI and Anthropic revenues alone have gone from 0% of GDP to ~1% on a run-rate basis in 18 months. SF1 expects 4 portfolio companies under 12 months old to cross $100M ARR this year. Valuations for the fastest-growing AI companies are cheaper than mediocre 2021 SaaS.

NCT

OKRs are sooooo old school. Ravi Mehta's NCT (Narrative, Commitments, Tasks) model simplifies goal setting. So instead of vague objectives, start with a clear Narrative explaining the "why" behind each goal. Next, set 3-5 measurable commitments for the quarter, with Tasks as actionable steps. The difference is that OKRs can be overly ambitious, but NCTs focus on achievable milestones that closely align with strategic priorities, more Agile in a way, as course corrections are easier (and it increases team accountability).

MARKETS

This is a densely packed summary - but with the top 10 private market companies, five or is it six (SpaceX and x.ai are counted separately) on the top of that list are now AI-first. Top 10 EV/NTM revenue multiples re-rated to 19.3x avg (16.3x median), up from 16.1x in April - jesus-bubbling-christ! M&A deals hit record average size ($260M) but at lower multiples. The 2022 correction looks complete - just not that evenly distributed if you ain't AI.

FOMO

"My boss brags it will 10X me." - every time I hear this, a GPU dies and goes to silicon heaven. The Walrus editor (Carmine Starnino) spent a week with people being told to use AI to do their jobs. He found a bunch of pressure to adapt without a plan or any rationale. People adopting shit that no one fully understands- anyhow - he found a third state between believer and skeptic: "collective acquiescence" - a surprise entry to our tech dictionaries.

DELEGATED BUYERS

Bessemer has an eComm reframe for us all: brands now need to optimize two front doors. One for AI agents parsing structured signals - clean product data, transparent pricing, accurate inventory. The other is for a human actually walking through. Impulse purchases may be an endangered species, check this quote from the article: "The buyer arrives better informed than the seller." FYI - 15-20% of retailer referral traffic now comes from AI chat. McKinsey projects $1T+ orchestrated US B2C retail by 2030, $3-5T globally.

CONVERSATIONAL ADS

A fast follow from above on retailer traffic (that makes me cringe a bit) - 460M people/mo now discover products inside AI assistants. Similarweb's read in this report: 46% of ChatGPT users who opened with zero commercial intent developed buying signals before an ad appeared. 83% of ad-triggering queries would never have triggered a Google Shopping ad. Conversations don't replace the intent; they generate it.

GOOGLE

You may have noticed Google Search being a little different this week - just got rebuilt for the first time in 25 years- here is why (and more) - at the Google I/O Conference last week, the CEO announced 2030 as the year we get AGI - he also made 24 other announcements in 35 minutes, AI glasses ordering DoorDash (live on stage), background agents inside Search and a universal cart across Search, Gemini, YouTube, Gmail (see #7 and #8 above for more)

CASE STUDY

OpenAI's enterprise lead over Anthropic collapsed from 41 points to 8 points in 12 months. Claude went 21% to 48% share. Gemini 27% to 40%. OpenAI peaked at 62% in Sep 2025, down to 56% - the first YoY drop. ETR pins the shift on coding assistants. Grok is still a rounding error at 7%

Issue 373

Permalink →

SaaS METRIC OF THE WEEK

CARR - Contracted Annual Recurring Revenue. This is a forward-looking SaaS revenue metric that estimates the maximum revenue size of a SaaS company, measuring current recurring revenue from your SaaS P&L and future revenue that sits in newly won customer contracts.

USER LED GROWTH

ULG is when your existing users become your biggest advocates, driving leads straight into your funnel. It's not for every B2B SaaS company, but when done right, it can build a flywheel that slashes your CAC and ramp times. A classic example is Dropbox's referral program, where users earned extra storage by inviting others—a powerful way to turn customers into evangelists one GB at a time.

SEED

Thinking Machines raised $2B at a $12B valuation and called it Seed. Humans& closed $480M Seed, and Periodic Labs took $300M. Meanwhile, the median US Seed round sits at $4.1M, almost exactly where it sat five years ago. Same stage by name, but it ain't the same market.

PRE-SEED

Carta's State of Pre-Seed: Q1 2026 report is out - AI startups now capture 50% of all pre-seed dollars on Carta, up from ~30% a few years ago. That's the same concentration you see at later stages - the filter is happening at the very top of the funnel. Total pre-seed capital is stable at $2.5-3B per quarter. The pre-seed middle is hollowing out. Rounds between $1M and $2.5M dropped from 24% of all pre-seed deals in Q1 2023 to 18% in Q1 2026. Sub-$1M rounds and mega-rounds both gained share.

OUTPUTS

I bet somewhere someone is asking, "Hey! where's the ROI on the 2M euros we paid Anthropic last year?" Robert Glaser argues most companies are measuring AI wrong - counting seats, prompts, PRs. The better question: what loops closed faster, which decisions improved, which product ideas got killed earlier because a prototype made the weakness obvious?

SKILLS

Skills are the new prompts engineering - Anthropic, OpenAI, Cursor, Microsoft, Vercel, Windsurf, and Lovable have all converged on skills as a shortcut for context that everyone appreciates. MCPs took 2 years to achieve something that took Skills 6 months. BTW - WTF, Lovable is at 400M ARR, which is $2.77M revenue per head.

DISCOUNTS

Lemkin's call: the best AI-native B2B companies stopped closing with "20% off if you sign by Friday" and started closing with "we'll deploy the agent for you this week, no cost." Discounts ask the buyer to make a bet. Deployment removes the bet. Once an agent is live, it's almost impossible to rip out.

CASE STUDY

In July 2025, aReplit agent deleted SaaStr's production database(and during an explicit code freeze), fabricated 4,000 fake users to cover its tracks, then admitted "I violated explicit instructions, destroyed months of work." Since then: 8 documented disasters. AI now generates code faster than humans can review it.

▌ Older issues Full archive (376) →

Get the top 10, every Friday

Curated SaaS and tech insight from around the web, repackaged for people to put to good use — free.