Nobody Buys Software Anymore. They Describe It.

The change underneath the noise

Every vendor who has called you this year has said the same thing: "good news, our product has AI in it now." You have heard it six times and believed none of it. Underneath the noise is a real change, and it is not about tools.

Here is the proof. Early this year, Anthropic shipped software that does office work - reads, writes, updates records, finishes the task. In the first days of February, an analysis published in Forbes counted roughly $300 billion in software market value evaporating in a single session. By the end of the month, nearly a trillion dollars had been erased from software and services stocks. A Jefferies analyst named it the "SaaSpocalypse" in comments to Bloomberg, and the name stuck. Salesforce and Workday slid hard. When Anthropic demoed legal work in its agent, even the software-sector ETF dropped.

The market was not reacting to a product. It was reacting to a fact your CFO already suspected: the cost of making software has collapsed, and every software bill you pay was priced when that wasn't true.

What you were actually buying

For thirty years, your company has rented software. But "software" was never one thing. It was four things sold as one:

1. The code. Someone wrote it so you didn't have to.
2. The encoded practice. The vendor's opinion about how this workflow should go - learned from a thousand customers and baked in.
3. The operations. Hosting, uptime, patching, security, backups. The boring machinery that keeps it running at 3 a.m.
4. The accountability. A throat to choke. An SLA. A name to put on the line when a customer's security questionnaire asks who is responsible.

Here is the whole argument in one move: AI collapsed the price of #1 and a good deal of #2. It did approximately nothing to #3 and #4.

Every prediction that SaaS is dead ignores 3 and 4. Every defense of the status quo ignores 1 and 2. The truth is a redistribution, not an extinction - and knowing which half you are paying for is the actual executive decision.

This is already happening, and not just at startups

You might assume this is a Silicon Valley story. It isn't. The evidence is in the middle market, and most of it is not coming through your IT department.

• A Retool survey of builders this spring - covered by VentureBeat in April - found 78% of teams planning to build custom tools by the end of 2026.
• Klarna, the payments company, ditched Salesforce's flagship CRM product in late 2024 for a homegrown AI system. Salesforce and Workday have been sliding ever since as investors realize other companies can do the same.
• The trade press reports CIOs and CFOs pushing back on "license sprawl" - dozens or even hundreds of overlapping SaaS tools bought over the years through decentralized departmental purchases, most of them with seats nobody uses. Renewals that used to be automatic are now being renegotiated, downsized, or postponed.
• The market has repriced the whole category: forward earnings multiples on software companies were cut roughly in half in a matter of months, from about 39 times earnings to about 21.

The number to dwell on is not any of these. It is the pattern underneath: most of the building is happening quietly. Tools get assembled by the people who actually do the work - a commission sheet here, a reconciliation helper there - and nobody in IT is asked, because nobody in IT was ever going to say yes. One Zoho CEO called AI "the pin that is popping this inflated balloon." What he is describing is not a story about IT budgets. It is a story about people building things without asking. That is the actual subject of this article.

The precedent nobody mentions: the spreadsheet

Everyone can already build software. They have been able to since 1985. It is called Excel, and your company runs on it.

Somewhere in your building is a spreadsheet that decides pricing, or commissions, or capacity. It was built by someone in operations, not IT. It has no version control, no tests, and no documentation. If that person left, it would take a month to reconstruct. Nobody calls it software. It is software - mission-critical, written by a non-programmer, and it exists because buying a product to do that specific job was never worth it.

Two lessons, and you need both.

It works. The spreadsheet is arguably the most successful end-user programming environment in history. It exists precisely because the last mile of every business is too specific to buy off a shelf. Your commissions spreadsheet is more accurate to your business than any vendor's commissions module, because it was written by someone who had to live with your actual rules.

It is also where the quiet disasters live. Unversioned, unreviewed, dependent on one person. The famous public spreadsheet failures - a trading desk, a published macroeconomics paper - were not caused by stupid people. They were caused by software that nobody treated as software.

AI is Excel's second act, with a far larger blast radius. Same democratization, same upside, same failure mode - now applied to things that touch customers and money directly.

What actually happens to the department stack

Here is the shape of the change: a department's twelve tools collapse into one system of record plus a thin layer of self-authored surfaces.

The system of record survives, and gets more valuable. It holds the data of record, the audit trail, the compliance burden. Nobody is rebuilding their general ledger, and nobody should. The workflow layer - the point solutions, the dashboards, the approval flows, the "we bought this because the CRM couldn't do X" tools - gets built in-house, used for a quarter, and thrown away.

Walk it through departments you recognize:

• Sales and marketing: keep the CRM. The six point solutions bolted around it become a handful of internal surfaces.
• Finance: keep the ERP. The reporting layer, the reconciliation helpers, the variance memos - internal.
• Operations: often keeps almost nothing. A database and a set of purpose-built screens.

Name the concept: disposable software. Software with a lifespan measured in weeks, written for one team and one quarter, then discarded without ceremony. Software stops being a capital asset you procure and amortize, and becomes a consumable you use up. That is the actual shift, and it is stranger than "everyone is a developer."

Here is the punchline for a company your size: for the first time, a 60-person company can run software as specific to its operation as a Fortune 500's - without a Fortune 500's budget. Specificity used to be a luxury good. It is not anymore.

The new bottleneck

When implementation is free, the constraint moves. It moves to the person who can say precisely what they want and iterate on the answer.

Name it plainly: specification. The imagine, articulate, correct, repeat loop. What exactly does this screen show? Who is allowed to change it? What counts as done? The tool used to answer those questions for you, because the vendor baked an answer in. Now the questions come back to your building.

Three things to understand about specification:

• It is not a technical skill. It is the same skill as writing a good brief, a good job description, a good SOP. Which means the person best positioned in your building may be an operator, not the IT contractor.
• It is rare, and its rarity was invisible until now - because until now nobody was ever asked to be precise. Buying software was a way of outsourcing the thinking about how work should go. That outsourcing is what is ending.
• It is learnable. That is the only honestly optimistic claim in this piece.

The uncomfortable corollary: a lot of organizational vagueness has been hiding behind software purchases. When the tool stops making the decision, the company has to. Some companies can't.

What breaks

The LinkedIn version of this story ends here, with a fist pump. The real version does not. Six honest failure modes:

1. The maintenance cliff. You built it, you own it - forever, including the 2 a.m. version.
2. The verification gap. You can now generate code faster than any human can review it. Generation was never the bottleneck people thought it was. Correctness was.
3. Shadow IT at industrial scale. People building without asking is not empowerment, it is an unmapped estate. You cannot secure what you don't know exists. When Retool shipped an enterprise-governance product in June, the hook was that "vibe coding" - AI-assisted software written by non-developers - now tops the C-suite list of concerns.
4. Data gravity. The tool is the easy part. The plumbing - permissions, sync, reconciliation - is where the months go.
5. Liability. No vendor to sue. No SOC 2 report to hand your enterprise customer. For regulated industries, this alone is decisive.
6. The incumbents are not standing still. Salesforce reported a strong quarter in February - $10.7 billion in revenue, up 13% year over year - and its CEO Marc Benioff said the word "SaaSpocalypse" at least six times on the call. "You've heard about the SaaSpocalypse? And it isn't our first. We've had a few of them," he said. Its agent platform, Agentforce, passed $800 million in annual recurring revenue, up 169% year over year. ServiceNow - which positions itself as the "AI control tower" for agent orchestration and ranks first on the SaaSocalypse public-market tracker - had its CEO tell customers at its May conference that AI intelligence is commoditizing. And the model makers are racing to own the layer above the software: OpenAI's enterprise agent was itself the trigger for a second February sell-off. The point is not that the giants are doomed. It is that everyone - vendor, model maker, and you - is fighting over the same question: whose system decides how the work actually runs.

Concede this properly, because it makes everything else survive contact: what is dying is not software-as-a-service. It is the per-seat pricing model and the thin workflow layer. Systems of record, infrastructure, and anything carrying compliance weight get more valuable, not less.

Five moves to make this week

1. List the friction. From the record, not from memory: the calendar, the sent folder, the file with the most "_final_v3" copies. Or run the meeting, if you have a team. Where is the day to day annoying in a way it should not be?
2. Rank what each friction really costs. Tools, people, who it is accessible to, and the hours or dollars. One line each. The ranking is the map.
3. Pick one test pilot. It must touch money or customers, repeat at least weekly, have rules that can be written down, and produce output that can be compared before it acts. Small enough to finish.
4. Name the owner and write the page. One person accountable. Then one page answering four questions: what it shows, who can change it, what counts as done, what happens when it is wrong. A cold reader should be able to answer an invented case from the page alone.
5. Set two rules, then build the ugly version in days. Where the data lives, and who reviews the output. Then build the smallest thing that runs against real work. The learning is the deliverable, not the tool.

The question changes

Call it disposable software, or the specification economy - name the behavior, not the category, because categories get claimed by whoever has the marketing budget and behaviors get quoted. Software stops being a capital asset you procure and becomes a consumable you direct.

The question at the center of your company stops being what should we buy and becomes what exactly do we want - which is a harder question, and a better one.

I'm talking to fifteen presidents about this right now, mostly companies between 50 and 500 people. If you've built something internally - or tried to and stopped - I'd like to hear about it.