For decades, software was expensive to build and cheap to deliver to one more customer. That combination produced a business model investors valued highly for two decades. A company invested heavily up front, sold the same product many times, and collected recurring revenue that grew with its customers.
AI changes the first half of that equation. Software is becoming cheaper to build, and increasingly cheaper for a competitor to replicate. The costs of acquiring customers, understanding their businesses, maintaining their systems, and accepting responsibility for outcomes have not fallen nearly as fast. The barriers to creating software are falling faster than the barriers to building a durable software business.
That does not mean the software industry disappears. It means the ability to write software becomes a smaller part of what makes a software company valuable. This series asked who captures the surplus when that happens. Each piece answered from a different seat.
What the series found
A profitable software company can lose value because investors doubt how long its revenue will last. Median software margins rose while multiples fell, and technology went from a premium to a discount in middle-market PE deals. Higher interest rates and slower growth explain part of that decline. AI adds a new risk to future pricing power on top of that correction, although the deal data alone cannot say how much of the decline it explains. (What Got Repriced)
An enterprise can replace an incumbent's interface and become more dependent on its infrastructure. Buyers are rebuilding point tools and putting agents on top of their systems of record. Each integration that lets an agent reach the vendor's data can deepen the lock-in it was meant to escape, and the savings from cheap building can leak to model providers, bundled AI, and maintenance. (What the Corporate Buyer Decides)
A vertical SaaS vendor can keep its customers and its data while losing the value of the workflow. Enterprises are already putting agents on top of their systems. If those agents reduce paid seats or take over billable workflows, the renewal could shrink even though the customer never leaves. No broad data yet show that happening across many companies, so it is a risk to watch more than a measured trend. Survival and value capture are different questions. (What Still Stands in the Middle)
A software startup can be an excellent business without being a good venture investment. Cheap software may make narrow markets viable, while venture funding concentrates in a small number of large bets. Software entrepreneurship and venture-backed software entrepreneurship may be diverging, although the evidence so far is preliminary. (The Squeeze on Software Startups)
Selling the finished job opens a larger budget without securing it. The world pays far more for work than for software, and providers that take responsibility for completed work can compete for that budget. Their margins and valuations may look more like a services firm's, and per-job prices are likely to fall as providers multiply. The value lasts only where the provider owns something rivals cannot reproduce cheaply. (Selling the Finished Job)
The economic benefit of cheaper software does not necessarily go to whoever writes the code. Across the five pieces, the pressure points toward customers who can bargain, model and cloud providers who bill for usage, companies that sell finished work, and systems of record that agents have to consult. Vendors whose value sat in features that became cheap to build are likely to keep less of it. Value moves toward whoever owns what stayed expensive: the customer relationship, the trusted record, or responsibility for the completed work. Most of the evidence so far shows where that pressure is building, and how much each group keeps will show up in margins, prices, and revenue as contracts renew.
What would change the picture
The shift is a structural transition with several plausible outcomes, and the series describes the direction of travel as the evidence shows it in October 2026. The scenarios below are conditional. Each would favor different participants, and each has a signal that would show whether it is happening.
Future development | Likely beneficiaries | Who loses ground | Signal that would show it |
|---|---|---|---|
Maintaining internal software proves expensive | Established vendors | Internal build programs and AI-native challengers | Spending on maintaining internally built applications |
Agents cut paid seat counts across many customers | Enterprise buyers and agent platforms | Seat-priced incumbents | Seat counts and net revenue retention at renewal |
API access becomes a valuable metered service | Systems of record | Buyers and agent layers built on top | Vendor pricing for data and agent access |
Model inference costs fall quickly | AI application providers and customers | Model providers' margins | Enterprise spending on models and inference |
Small, specialized markets become economical | Bootstrapped and independently owned software businesses | Venture-backed horizontal tools | Survival and profitability of small software companies |
Enterprise AI usage concentrates among a few providers | Large platforms and infrastructure suppliers | Buyers' bargaining power | Share of enterprise model usage held by the top providers |
Incumbents move quickly to agent-based pricing | Established vendors | AI-native challengers | Incumbent pricing changes at renewal |
Finished-job pricing spreads across services workflows | Providers that can carry liability for completed work | Firms that bill by the hour, and seat-priced vendors | Published per-job prices and services firms repricing from hours |
Per-job prices fall toward the cost of the work | Buyers | Providers without regulated authority, a trusted record, or risk capacity | Per-job prices over time as providers multiply |
Two further developments would cut across all of these. New forms of defensibility could appear, as they have in past shifts in software economics, with agent compatibility, trusted evaluation, and accountability for automated decisions as early candidates. And the correction could prove mostly cyclical. If growth recovers and rates fall, part of the repricing could reverse, as the partial rebound in software stocks since April 2026 suggests. The test is whether the recovery reaches the typical vendor or stays concentrated among a few.
Signals worth watching
These signals would show which way the surplus is moving. Each is tagged with the stage it measures, since most of the evidence so far covers cost reduction and bargaining. A follow-up to this series will report what they did.
Signal | Stage | What it would indicate | Where to look |
|---|---|---|---|
Tech multiples relative to all industries in middle-market PE deals | Realized value | Whether the discount to other sectors persists | |
The gap between the top and the typical public software multiple | Realized value | Whether value is concentrating in a few companies | |
Seat counts and net revenue retention at renewal | Realized value | Whether workflow value is leaving vendors that keep their customers | Public SaaS earnings |
Spending on maintaining internally built applications | Cost reduction | Whether buyers keep the savings from building | Gartner and enterprise surveys |
Enterprise spending on models and inference | Realized value | Whether infrastructure providers capture the surplus | |
The share of new companies led by solo founders or financed without venture capital | Cost reduction | Whether cheaper building is changing who starts software companies | |
Survival and profitability of small software companies | Realized value | Whether narrow software businesses last | Founder surveys and acquirers of small software companies |
Incumbent pricing changes, such as retiring seats or bundling AI | Bargaining | Whether incumbents are adapting or extracting | Vendor announcements and renewal terms |
A sustained wave of software companies acquiring regulated service providers | Bargaining | Whether vendors are taking responsibility for results | Deal announcements and regulators' registers |
Human minutes per job, exception rates, and gross margin at AI-native service companies | Realized value | Whether finished-job revenue carries software or services economics | Company disclosures and investor updates |
The question that remains
The consequences of cheaper software are already visible in public-market valuations, enterprise procurement decisions, PE portfolio strategies, and startup funding. What comes next is not necessarily fewer software businesses. It may be more of them, built with less capital, serving narrower markets, and capturing value in ways that look different from traditional SaaS, including by selling the finished job instead of the tool.
The question is no longer only who can build the software. It is who owns the customer relationship, who accepts responsibility for the completed work, and who keeps the economic value.
Read the series
What Got Repriced (The Market): Why profitable software companies lost their valuation premium, and who keeps the surplus when code gets cheap.
What the Corporate Buyer Decides (Enterprise): How corporate buyers weigh economics, control, and accountability when deciding what to build, buy, and renegotiate.
What Still Stands in the Middle (Mid-Market): How vertical SaaS vendors can keep their customers and still lose the value of the workflow.
The Squeeze on Software Startups (Startups): Why software entrepreneurship and venture-backed software entrepreneurship may be diverging.
Selling the Finished Job (Services): When the money moves from the software to the work it does, and what keeps it there.
Sources
Sources for each finding are listed in the individual pieces. Sources cited directly here:
Middle Market Transaction Update, Fall 2026, Mercer Capital
The SaaSpocalypse Is Over, SaasRise, citing Meritech, July 2026
2025: The State of Generative AI in the Enterprise, Menlo Ventures, December 2025