6 min read · 13 sources

Miro and Airtable both got cash flow positive, and both sold this month for about 2.5x revenue. Between them they held roughly $1.4B in cash, and the buyers paid nothing extra for a cent of it. Growth rate was the only thing priced. If you are hoarding cash instead of investing it in growth, the market just showed you what that is worth: 1.0x.

That is the story that ties together the biggest news of the day, and it lands alongside Paul Graham telling founders to stop optimizing for profit and start optimizing for power. The two arguments are the same one from different directions. Here is everything else that matters.

Between them, Miro and Airtable held about $1.4B in cash, and buyers paid nothing extra for it - the cash sat there earning the worst multiple on the cap table.

Miro and Airtable Prove Cash Flow Positive Buys Nothing

Source: saastr.com ↗

Bending Spoons closed two acquisitions this month: Miro for $1.355 billion and Airtable for $1.285 billion. The breakdown is brutal for anyone who has been told profitability is the goal. Airtable had about $480M in ARR growing over 20% and sold for roughly 2.5x revenue. Miro had about $600M in ARR and sold for $1.355B - also about 2.5x. Miro’s announcement did not even include a growth rate, which tells you what the seller thought of that number.

The cash is the kicker. Airtable’s purchase price was 43% cash already on the balance sheet; Miro’s was 24%. A growing company’s ARR priced at 2.3x-2.7x in these deals, versus 8x-10x for a company growing 60%. Cash is valued at 1.0x in an acquisition. Hoarding it to reach profitability is the most expensive thing a startup can do.

Paul Graham: Build Power, Not Profit

Source: paulgraham.com ↗

Paul Graham’s latest essay argues the most useful heuristic for a startup is not “how do we make more money” but “what would make this company more powerful?” The former leads to incremental improvements; the latter can produce order-of-magnitude jumps.

His concrete moves: own the customer relationship, make money flow through the company, and build app-store-like platforms or network effects even in unexpected places. The “deluxe version” is a full app store, but simpler network effects work - let users share data or opt in to training AI models. He also suggests generalizing an idea, like letting AI agents pay each other, to turn a service into a marketplace. For engineers, the reframe matters: Graham is describing architecture decisions - who owns the data, who owns the transaction - as the core of strategy.

Conway's Law Now Includes the AI in Your Org

Source: domenkozar.com ↗

Conway’s Law - software mirrors the communication structures of its builders - is not dead, but the structures have changed. AI is now a core part of them, and workflows have to adapt or drift toward one extreme.

The evidence is in the tooling. Projects like Ghostty and Zed have added instructions or markers specifically to catch AI-generated contributions that lack human inspection - the author compares them to school tests designed to catch students who did not read. The tension is structural: demanding detailed evidence (tests, benchmarks) for AI contributions is expensive for human contributors, but the volume of AI-generated work can overwhelm human review. Projects must evolve their trust mechanisms to accommodate both, or they will drift toward being predominantly one or the other.

LangChain Open-Sources Its Paid Media Agent

Source: langchain.com ↗

LangChain released the Paid Media Agent that scaled its marketing from organic to five paid channels in six months. The numbers: paid media grew from 0 to 20% of marketing pipeline, and cost per qualified lead dropped 30% from June to August while spend rose 60%.

The technical details are the useful part. The agent tracks product announcements, drafts campaigns, adds keywords, and tests variations as a continuous learning loop. The system prompt acts as a map so the agent finds tools without carrying everything in context - “use models for judgment and code for consistency.” The hard problems were reconciling different ad platform data schemas and mapping campaign parameters to business outcomes. The agent proposes changes, routes them through human approval, then verifies they were applied.

Slackforce Surfaces: Dashboards From a Prompt

Source: slack.com ↗

Slack launched Slackforce Surfaces, letting any user build live interfaces like dashboards, reports, and polls directly from a Slackbot prompt, grounded in Salesforce and other connected systems. The interfaces show their data sources, can stay live and auto-update (coming soon), and require no analyst or designer. Pin one to a channel and the team can filter, comment, and act on it together.

This is the Slackforce integration becoming real for non-power-users. The pitch is that AI-powered reporting stops being a specialist tool. For anyone who has built a dashboard pipeline, the interesting claim is the grounding: visible sources mean the output is traceable.

PostHog's 100 Events a Year: Don't Ask Engineers to Demo

Source: posthog.com ↗

PostHog scaled from 1 to over 100 IRL events in a year with over 50% of the company participating in demos. The secret is that they never ask. The engineer who built a feature demos it, the travel budget and brand assets are provisioned, and nobody needs approval.

The mechanism is hiring “product engineers” who own product decisions, pricing, and customer support - so sharing their work publicly is a natural extension of the job, not a favor. About 95% of those events put engineers face to face with customers, and about 20% of engineers opt out with no consequences. The contrast is with companies where speaking is treated as a reward or obligation.

What Bending Spoons' Takeover Actually Looks Like

Source: alexkras.com ↗

An engineer who lived through it wrote up the six months after Bending Spoons acquired Evernote, and it reads as a preview of Miro and Airtable’s futures. Evernote was around $100M in ARR when 129 people were laid off in one round. His team went from six developers to one. The subscription price went from $69.99 to $129.99 a year. Almost everyone was eventually laid off; he quit on his own terms and is one of the few who can talk about it, because severance packages carry non-disparagement clauses.

The detail that matters for anyone at an acquired company: he reads a voluntary severance offer as the new owner announcing its plan for the org. Bending Spoons later listed at $18.4B.

The Rest

Source: cruciblecapital.substack.com ↗

The “compounding capital” thesis argues capital intensity is now a core moat, and founders are building two products: the one sold to customers and the “risk” sold to capital. High valuations make equity dilution cheaper than debt, per the Miller and Modigliani framework.

This LinkedIn ads playbook (sponsored) says LinkedIn only pays off when deal size is at least $10K and there are 10,000+ decision-makers, with $3,000 a month as the minimum test budget. One account switching to manual bidding took cost per lead from $287 to $90.

This PLG sales post (sponsored) names the “original sin”: “Why do we need 10,000 $10 customers when we could get one $100K customer?” The math works in Excel but ignores that the $100K customer usually started as a $25 self-serve user.

Marty Cagan admits he “completely understated the importance of business viability” in the first edition of INSPIRED, and that product risks were only value, usability, and feasibility - viability buried under feasibility. AI products make that worse with cost, monetization, liability, and ethics risks.

This mentorship defense argues LLMs produce better code but do not make the programmer better at writing it - a mentor can observe the flawed reasoning behind the code, not just the result.

And a16z argues the biggest VC mistake is omission, not commission - passing on SpaceX, OpenAI, or Stripe. With SpaceX public at ~$2T, and Anthropic and OpenAI rumored to go public at ~$2T and $852B, historical asset allocation models are obsolete. One winner, like Databricks, can represent 20% of a firm’s total AUM.