Three posts ago, the question was: does software lose value when AI can write it faster? The Q1 earnings gave us the data answer: yes, for seat-count models tied to headcount. But in March, the institutional investors started publishing their theory answers — and they’re more interesting than the data.
Two essays landed within weeks of each other. They’re from firms that rarely agree on framing. On the surface they look like contradictory takes. They’re not.
Sequoia: sell the work, not the tool
Sequoia’s “Services: The New Software” opens with a claim sharp enough to be falsifiable:
The next $1T company will be a software company masquerading as a services firm.
The argument is built on a single observation: for every dollar spent on software, six are spent on services. SaaS has been capturing the tool budget. The autopilot generation captures the labor budget.
The framing they introduce is worth holding onto: copilots vs. autopilots. A copilot sells the tool. An autopilot sells the work. Harvey sells to law firms (copilot). Crosby sells to the company that needs an NDA drafted, not to outside counsel (autopilot). The difference isn’t just go-to-market — it’s which budget line you’re displacing.
Copilots make professionals more productive. Autopilots eliminate the professional from the transaction.
The essay maps every major outsourced services vertical on two axes: intelligence-to-judgement ratio (how automatable is the core work?), and outsourced-to-insourced ratio (is there already an external budget to displace?). The highest-priority targets are highly automatable work that’s already outsourced: insurance brokerage ($140–200B), healthcare revenue cycle ($50–80B), claims adjusting ($50–80B), accounting ($50–80B).
The strategic logic is clean: outsourced work means the company has already decided this can be done externally. There’s already a budget line. The autopilot is a vendor swap, not a reorg.
The flywheel: start with the outsourced, intelligence-heavy task. Nail distribution. Compound proprietary data. Expand into the insourced, judgement-heavy work as the AI gets smarter on your domain. The outsourced task is the wedge. The insourced work is the long-term TAM.
a16z: the moats aren’t going away
a16z’s response to the same moment is different in tone but compatible in logic.
“Good News: AI Will Eat Application Software” opens by acknowledging the SaaSpocalypse directly: IGV down 30%, all gains since ChatGPT erased. Then it makes a structural argument: the market is mispricing what software companies actually sell.
Code is never where the value has lived. If code is where the value was, these companies would have been killed years ago by open-source software or by cheap engineering labor in developing countries.
The four bear cases (foundation models go up-stack; enterprises vibe-code replacements; incumbents expand and collide; new entrants undercut on price) are real pressures. But a16z runs them through Hamilton Helmer’s Seven Powers framework and argues that the moats that actually matter — network effects, cornered resources, process power, brand — aren’t going away. The one moat that is weakening is switching costs, because AI makes migrations easier. Their counterintuitive take: that’s good for software, not bad. Companies that had “hostages, not customers” will finally have to earn their business.
The companion piece, “There Are Only Two Paths Left for Software”, drops the theoretical frame and gets operational. The message to software CEOs is direct:
The comfortable middle is over.
Two credible paths to durable equity value:
- Accelerate revenue growth by 10+ percentage points, year over year, through genuinely new AI-native products — within 12–18 months.
- Rebuild to 40%+ true operating margins (including stock-based compensation as a real expense).
Everything between those paths is “no-man’s land.” Growth pressure, persistent dilution, multiple compression. The essay is unusually specific: the “8–10% layoff” is the weak form. The strong form is a redesign of the machine.
Reading them together
The surface tension between these two essays is real. Sequoia says the next big companies will sell labor outcomes, not tools. A16z says the existing software industry will survive because its real moats were never in the code.
But these aren’t contradictory. They’re describing different segments:
- For new entrants: Sequoia’s frame applies. Build autopilots. Target outsourced labor budgets. Don’t compete with the tool; compete with the service firm.
- For incumbent SaaS companies: a16z’s frame applies. Your network effects and data moats are intact. But your switching-cost moat is eroding, and your cost structure needs to reflect that. Choose growth or efficiency. Not neither.
The synthesis: the software budget is being squeezed; the services budget is being opened up. Companies that can move from one to the other will compound. Companies stuck selling tools into a world where tools are cheap face the “two paths” ultimatum.
How this updates our hypothesis
We’ve been tracking: software loses value when AI reduces the number of humans who need to use it.
These essays sharpen that into something more precise:
- Seat-count models compress when AI reduces headcount. We have the data on this from Q1.
- Tool-as-product models compress when the model does what the tool did. Sequoia’s “copilot race against the model” is exactly this.
- Work-outcome models expand when AI gets better, because every improvement makes the service faster, cheaper, and harder to compete with.
The question isn’t just “does software have value?” It’s: which kind of value claim is your company making? Tool access? Seat-based workflow management? Or a guaranteed outcome?
The companies that can answer “outcome” — and back it with proprietary domain context accumulated over years — are the ones the Q1 compression isn’t touching. Palantir at 70% revenue growth. Snowflake at 30% full-year revenue growth. They aren’t selling seats. They aren’t selling tools. They’re selling things you can’t replicate by spinning up a cheaper model.
That distinction is now the load-bearing variable in software equity value. We’ll be measuring it against Q2 data when it arrives.
— Chris
Sources: Sequoia Capital — Services: The New Software · a16z — Good News: AI Will Eat Application Software · a16z — There Are Only Two Paths Left for Software