8 min read · 14 sources

AI startup ARR is a six-month rental. Madrona surveyed 150 enterprise IT buyers and found 77% reevaluate their AI vendors every six months or on a rolling basis, with 74% planning to expand AI budgets in the next 12 months. The number that should scare founders: fewer than half of AI pilots ever make it to full production. Even when they do, the customer keeps shopping.

That is a different shape than SaaS. Traditional enterprise software locked in on multi-year contracts and integration cost; AI vendors face “fast in, fast out” dynamics where a buyer can swap providers between budget cycles. A separate Andreessen Horowitz survey of 50 technical AI buyers found more than half want AI fees tied to outcomes rather than tokens or seats, which is a pricing problem most startups have not figured out how to solve. For engineers, this means the ARR number your competitor just announced is closer to a quarterly rental than the durable revenue base it would have meant in 2021.

Only about 50 funds closed today can lead typical Series A rounds, versus roughly 200 a decade ago, as the median check quadrupled to $19.4M.

The scaling versus profitability trade-off

Source: aswathdamodaran.substack.com ↗

Aswath Damodaran, the NYU valuation professor, argues that venture capital’s tilt toward scaling over profitability has become its weakest link. He frames the founder’s choice as binary: build a smaller, profitable business with minimal external capital and full ownership, or pursue a larger market with external funding that dilutes control and defers profit indefinitely. The piece was prompted by Vinod Khosla’s now-deleted tweet defending the scaling-first model.

The architectural consequences are concrete. A founder optimizing for cash efficiency makes different build-versus-buy decisions, hires a leaner team, and ships a narrower product. A founder optimizing for growth capacity over-invests in infra, accepts technical debt, and treats revenue as a marketing surface. Neither is wrong, but the trade-off should be explicit, not implicit.

The Series A is dead, long live the Series A

Source: jaydimonte.substack.com ↗

Jay DiMonte does the math on what happened to the round in between. Median Series A size grew from $4.5M a decade ago to $19.4M today, a 4x increase over ten years and roughly 2x over five. Under a model assuming a 20-company portfolio, a 70% lead check, 30% reserves, and 20% fee load, only about 50 funds closed today can lead a typical Series A round versus roughly 200 a decade ago. That is one quarter the rate.

The money did not vanish; it concentrated. Many firms pulled back to pre-seed and seed, and a handful grew seed vehicles to $250M-$500M (Pear, First Round, Lerer Hippeau, Primary, per DiMonte). For an engineer-founder, the practical consequence is fewer writeable checks at Series A, which means more competition for the same dollars, longer fundraising cycles, and a stronger push toward growth-stage metrics earlier in a company’s life.

Forward-deployed everything

Source: insights.euclid.vc ↗

Quinn Litherland, founder and CEO of Revin (AI voice and SMS for home services), describes how his forward-deployed engineers embed on-site in customer operations. Revin’s agents become top-performing reps on day one, and Litherland says the company generates over $5M in net new revenue per customer.

Litherland flags two failure modes for FDE-as-strategy. Most companies staffing it have “sales engineers dressed up,” not engineers who can actually ship against an unfamiliar codebase mid-deployment. And if every problem gets a custom solution, the roadmap sprawls and the product never compounds. The teams that work look like mini-founders: they handle implementation, debug customer workflows, run success, and feed product decisions back. If your AI deployment org looks like a consulting shop with a marketing page, it is probably already failing.

How to get a warm intro

Source: x.com ↗

Paul Klay is blunt that cold outreach does not work for fundraising. The best VCs source deals; they do not read inbound. He argues investors think in two modes: patterns (observed trends like YC founders returning 10x) and theses (fund-specific focus like AI infrastructure, hardware and silicon, or physical AI). The warmest intro runs through a portfolio company that has made money for the fund, which you can find via Crunchbase.

For engineers, the framing is useful even outside fundraising. Investors pattern-match the way interviewers and hiring managers do; understanding that helps you pick which doors to knock on and what to lead with.

GojiberryAI: $4M ARR, zero funding, one year

Source: x.com ↗

A 30-year-old founder wrote up how he took GojiberryAI from nothing to over $4M ARR and into YC without outside capital. The tactics are unglamorous and specific: he sold before building, closing the first $10K off a six-slide deck. His customer was “founders at 20-person SaaS companies about to hire their first SDR.” Outbound targeted people showing intent signals (engaging with competitors, changing roles, raising money), and he claims 25-40% reply rates against the 1-2% baseline for scraped lists.

He led with blueprints instead of calendar links, picked one channel until it worked before adding a second, talked to customers daily, and landed at $99 a month with a free trial. The engineering takeaway is not “skip the engineering” but “sell the smallest testable version before writing the first line.” Most engineer-led founders invert that order and pay for it.

Agent Auth: agents get identities

Source: workos.com ↗

WorkOS launched Agent Auth in early access. The product gives agents real identities in AuthKit, with short-lived, tightly scoped tokens minted per run. Every token identifies the agent, the user or org it is acting for, and the permissions it carries, and the session can be revoked instantly.

This addresses a gap most enterprise security teams have been quietly worrying about. The current patterns (long-lived API keys stored in a config file, or agents borrowing a user’s session and inheriting everything that session can do) do not survive a security review once the agent starts touching real data. If you are deploying agents into enterprise customers, this is the shape of the integration you will be expected to support within a quarter.

Voice-first Workspace

Source: blog.google ↗

Google is shipping three new voice-activated Workspace features this week, all powered by Gemini Audio models. Gmail Live is a conversational inbox (“what’s my flight’s gate number?”). Docs Live acts as a hands-free thought partner that pulls from Gmail, Drive, Chat, and the web. Keep Live turns voice dumps into structured notes. They are rolling out to AI Plus, Pro, and Ultra subscribers in Gmail and Keep, and to Pro and Ultra in Docs, with Workspace business editions to follow.

For engineering teams building on Workspace or competing with it, voice-first is now a real product surface inside the productivity stack, not a demo.

AI raises the quality floor, not the speed

Source: x.com ↗

Richard Tong’s writing experiment tracked 15 published posts a year from 2021 through 2026. The 10th-percentile post’s quality score rose from 2.59 to 3.81 on a 1-5 scale, almost double the gain at the 90th percentile. Line-level edits per piece stayed flat at around 136.

The honest reading: AI did not cut his editing time. It cut the number of bad drafts that survived to be published. For engineering managers rolling out AI tooling, this is the metric to watch. Time saved per task is a vanity number; raising the floor on what gets shipped is the durable one.

Acclimate before you fundraise

Source: chrisneumann.com ↗

Chris Neumann tells founders outside California to spend two to three weeks in San Francisco before starting a raise. The top 5% of seed valuations hit unprecedented highs in H1 2026 on the back of preemptive rounds, and the local tempo has gotten faster. Founders who show up speaking last quarter’s language and asking for intros read as “not going fast enough,” which is death in a competitive round.

Even founders with prior SF experience appear slow now. If you are remote and about to fundraise, budget the trip the same way you budget the deck.

AI is stopping startups from completing puberty

Source: ashley.rolfmore.com ↗

Ashley Rolfmore argues AI lets startups paper over capacity limits that would otherwise force structural decisions. She describes a team using an AI agent to correct customer output bugs their product could not fix; as the agent improved, the underlying product gap became invisible, so the crisis that would have forced a real fix never came.

This is worse when the product itself is AI, because customer interactions no longer feed back into product improvement in the way SaaS feedback loops used to. For engineering leaders, the takeaway is to instrument the things AI workarounds are masking, not just adopt the workarounds. Otherwise you wake up at Series B with a product-shaped object and no customer signal telling you why.

Solo founders dodge the biggest cause of death

Source: solofounders.com ↗

Citing Noam Wasserman’s Harvard research, Julian Weisser notes that roughly 65% of high-potential startup failures trace back to co-founder conflict. Solo founding trades that failure mode for isolation, which Weisser argues is “manageable in a way a bad co-founder isn’t.” He flags that “we don’t invest in solo founders” is often shorthand for “I don’t really believe in you,” and investors read founder judgment through hires; bring on three teammates and 75% of the company is no longer you.

Doug Leone, the Doug Leone playbook

Source: theaiopportunities.com ↗

Doug Leone, 69, came out of retirement after 26 years at Sequoia because he felt “good for nothing.” He gave himself a 9-month deadline to stay relevant. The summary of David Senra’s conversation covers Sequoia’s heuristics for picking founders, holding winners, and building boards: fear beats vision, drive comes from somewhere ugly, the three filters behind every check, why VCs sell too early, and Leone’s preference for “heart of a lion over resume.” The piece is gated.

Product or distribution: most of you have a distribution problem

Source: thefounderplaybook.hustlefund.vc ↗

A Hustle Fund analysis revisits Justin Kan’s 2018 tweet that first-time founders obsess over product and second-timers obsess over distribution. The fund estimates that around 5% of its early-stage portfolio has a real demand problem and the other 95% have a customer-acquisition problem. Product work feels productive and is within your control; distribution feels uncertain and scary. Engineers who recognize this in themselves ship earlier and spend more cycles on the channel, not the codebase.