Cheap code is not the same thing as cheap software service. That is the core argument in a BlockTempo report that turns to writer Joan Westenberg’s “bread paradox” to test one of the market’s loudest claims this year: that AI coding tools will wipe out SaaS.
The article revisits the software-stock panic from January, when the rise of Claude Cowork and Claude Code fed calls for a “SaaSpocalypse.” According to the report, software shares lost about $300 billion in a single day during that selloff. Its point is that cheap recipes and cheap machines did not destroy baking over the past 5,000 years, and cheap code may not do the same to software vendors now.
The bread machine analogy
BlockTempo frames the discussion with a household example. A bread machine can make one loaf in three hours, ingredients cost less than $3, and the machine itself costs about $100. On paper, that should make home baking an obvious choice. In practice, Americans still buy about 10 million pre-baked loaves a day.
Westenberg calls that gap the “bread paradox.” The issue is not whether people can make something themselves. It is whether doing so is truly cheaper once the hidden costs are included. The report ties this to the economic make-or-buy decision: people routinely underestimate the accumulated cost of small repeated tasks such as tracking ingredients, pressing the right settings, waiting through the process, and cleaning up afterward.
The article backs the analogy with a long historical arc. Commercial bakeries were already operating along the Nile around 3000 BCE. In Roman times, baking had become industrialized. By the time Pliny wrote Natural History, Rome had professional baker guilds, animal-powered kneading machines, and a logistics network that supplied bread to hundreds of thousands of city residents, most of whom did not bake for themselves.
It adds that London’s bakers’ guild had a royal charter by the 12th century, and bakers caught selling underweight bread could be tied to a sled and dragged through the streets. The pastoral picture of every household baking its own bread, the article says, tends to hold only when people cannot buy bread at all.
The pattern continued into the industrial era. Otto Rohwedder invented the commercial bread slicer in 1928, and the Chorleywood process in 1961 cut baking time from hours to minutes. Even now, Americans consume about 21 million tons of bread and baked goods each year and still buy around 10 million pre-baked loaves a day. The report’s takeaway is plain: low-cost inputs, affordable machines, and old know-how did not eliminate the commercial supply chain.
It also cites George Orwell’s complaint that industrial British bread was “pale, soft, and tasteless.” People kept buying it anyway, because buying bread carried a lower psychological cost than baking it.
AI lowers coding costs, not the cost of the full service stack
The article then applies the same logic to SaaS. The bearish case sounds compelling at first glance: with a strong enough AI model and a decent prompt, a company can generate a custom CRM or analytics dashboard in an afternoon. If code is nearly free and servers are cheap, why keep paying subscription fees?
BlockTempo’s answer is that companies are not paying Notion, Jira, or Basecamp for code alone. What they are buying is the accumulated output of thousands of engineers, compliance staff, and security auditors, along with institutional knowledge, integration ecosystems, regulatory certifications, and support infrastructure.
A company that builds its own system with AI, the report argues, is really buying the software equivalent of a bread machine. The ingredients may be cheap, and the machine may handle much of the work, but the business itself becomes the baker. It has to carry maintenance, edge cases, and the security risks tied to AI-generated code. The article cites research saying major defects appear in AI-generated code at about 1.7 times the rate found in human-written code.
It also points to the operational problem that often emerges later. Six months on, the employee who built the internal tool may have moved to another department. Nobody else fully understands how the system works, and when it fails at 2 a.m., no one is there to pick up the phone. Faster code generation does not solve that part of the job.
Renewal data suggests buyers are staying put
The report says the pricing and procurement signals point in the same direction. Gartner has observed that enterprise SaaS renewals have recently been rising by 10% to 20% in many cases, outpacing budget growth for most CIOs. That could look like vendors raising prices during a moment of confusion, but buyers do not seem to be leaving.
Avenir’s January 2026 report reinforces that reading. It found that 63% of enterprise buyers expect existing software vendors to benefit from generative AI, while only 8% expect them to be harmed. In other words, customers appear to want incumbent services to absorb AI capabilities, not to tear everything down and rebuild from scratch.
The article also addresses Klarna, which is often cited as proof that internal builds can beat SaaS. It says Klarna did not replace Salesforce by creating an entirely new system from zero with AI. Instead, it adopted a different mix of SaaS products and paired that with some in-house development. Its teams, the report adds, still rely on Slack inside Salesforce.
Thin products may struggle, but the supply chain remains
Where the pressure is real, the report says, is in thin products that can be copied with a single prompt. It lists examples such as converting PDFs into spreadsheets, generating meeting notes, or sending follow-up emails. Tools built around one narrow function were always on shakier ground.
That is different from SaaS companies with deep integrations, proprietary data, regulatory certifications, years of business logic, and partner ecosystems. Those businesses resemble the full industrial baking complex. Individuals can bake bread at home, but that has never been enough to threaten commercial baking at scale. Bakers do not sell flour and recipes. They sell consistency, reliability, and the assurance that someone is accountable when something goes wrong.
The next major shift, according to the article, is likely to be pricing. As AI agents become a new class of software user, seat-based charging may gradually give way to usage-based and outcome-based pricing. Thin, single-function products may disappear, and the report argues that they probably should. They were not durable businesses so much as artifacts of an era when software development was expensive enough for even trivial functions to carry monthly fees.
The article ends with a broader point about the staying power of SaaS. The real product is not code but a rented solution to an operational problem. That logic has held from the Roman Empire to the present. It survives not because the technical barrier is permanent, but because people will keep paying others to handle recurring headaches as long as the price feels reasonable and trust remains intact.