Key takeaways
- According to Google's 2025 DORA report, published on September 23, 2025 from a survey of nearly 5,000 professionals, 90% of software professionals now use AI, for a median of two hours a day.
- The Stanford study of roughly 100,000 developers measured a median productivity gain of 10 to 15%; teams that master the tooling reach 20 to 30%, while others gain nothing.
- The practical consequence for an SMB: a genuinely usable business application reaches production in 4 to 6 weeks, and a well-defined scope fits in a sprint starting at 400 euros (excl. VAT).
- AI speeds up writing code, not judgment: deciding what to build, and above all what not to build, remains human work.
Software development has just shifted gears, and that shift concerns small businesses directly. What used to put a custom business tool out of reach for a 20-person company, the sheer volume of repetitive code to write, is precisely what AI accelerates best. The useful questions are what it actually speeds up, what it does not, and how that changes the buy-or-build decision.
What the numbers say, past the promises
Google's 2025 DORA report, published on September 23, 2025, is the most complete snapshot to date: 90% of software development professionals now use AI at work, up 14% year over year, with a median of two hours of use per day. More than 80% of respondents say AI has improved their productivity, and 59% credit it with a positive effect on code quality.
The same report carries the counterpoint: 30% of the professionals surveyed have little or no trust in generated code, and only 24% trust it strongly. DORA sums AI up as an amplifier: it strengthens organizations that already have sound practices, and it magnifies the dysfunctions of the others.
The Stanford study led by researcher Yegor Denisov-Blanch provides the coldest measurement: on real data from roughly 100,000 developers, the median AI-related productivity gain sits between 10 and 15%, far from the 50% or 100% some vendors advertise. The spread is the real lesson: teams that master the tooling gain 20 to 30%, others gain nothing, and the clearest wins come not from writing code but from everything around it: understanding an existing codebase, debugging, verifying.
Why is custom software back within reach for SMBs?
Because the cost structure of business software has always been dominated by repetitive work: forms, list views, a back office, integrations between tools, tests. That is exactly the category of work where AI delivers its biggest gains. The scarce part of the craft, understanding a process and turning it into simple software, has not changed price; the voluminous part has shrunk.
I see it on my own projects. IA Brew, a fully automated newsletter, runs on 93 n8n nodes built and maintained solo. The back office behind mathieuhaye.fr tracks leads, SEO and articles without any off-the-shelf product. My portfolio monitoring dashboard relies on Claude Haiku 4.5 to produce a daily brief. And the article you are reading was written, translated and published by a custom-built bilingual pipeline. None of these tools would have been worth building at 2022 conditions; all of them now are.
For an SMB, that translates into simple public numbers: on my side, a well-defined scope fits in a sprint starting at 400 euros (excl. VAT), a genuinely usable business application reaches production in 4 to 6 weeks, and the quote is written before work starts. Three years ago, the same scope was measured in months and tens of thousands of euros at most providers.
What AI does not speed up
Three things resist, and they are what separates a tool that serves from a tool that gathers dust.
First, scoping: understanding why a process exists, what should stay manual, what deserves automation. No coding assistant asks those questions for you. Second, quality: generated code must be reviewed, tested and secured, and the 2025 DORA report notes that delivery instability remains a concern in teams that adopt AI without guardrails. Third, maintenance: software lives on, and a tool built fast but badly costs you every month that follows.
In other words, the cost of custom software has not disappeared: it has moved. Fewer hours of typing, just as many hours of judgment. That is good news for SMBs, on one condition: that the provider bills for judgment, not volume. The table below sums up where AI actually changes the equation.
| Type of work | Effect of AI | What it means for an SMB |
|---|---|---|
| Repetitive code (forms, back office, integrations) | Strong acceleration | A usable business tool in 4 to 6 weeks |
| Scoping and understanding the business | Close to none | The scoping call remains the decisive step |
| Review, tests, security | Load shifted, not removed | Demand it in writing in the quote |
| Maintenance and evolutions | Moderate acceleration | A light support plan is often enough |
When custom is still the wrong answer
AI lowered the cost of building software, not the cost of owning it. Three situations where I advise against custom, even now that it is affordable.
If your process is standard, invoicing, payroll, classic customer support, an off-the-shelf product covers it better and cheaper, subscription included. If your volume is too low, ten operations a week do not justify a dedicated tool: a well-kept spreadsheet does the job. And if your real bottleneck is elsewhere, commercial or organizational, the budget is better spent there. On the Horus Condition Report engagement, the right answer was a well-configured Pipedrive, not a CRM built from scratch; recommending it was part of the job.
The 10 to 15% median gain measured by Stanford looks modest; concentrated on the most expensive part of a business tool, it nonetheless changes the answer to a question many business owners had shelved: "is software of our own worth it?". I build the software your company wishes it had found on the market. If a manual process keeps resisting your current tools, the simplest step is to look at it together: book a 30-minute call, or browse the custom web app page to see how I work. If an off-the-shelf tool is enough, you will leave with its name.
Frequently asked questions
Does AI make custom software cheaper for a small business?
Yes, mostly on the repetitive code that dominates a business tool: forms, back offices, integrations between tools. The Stanford study of roughly 100,000 developers measured a median productivity gain of 10 to 15%, rising to 20 to 30% for teams that master the tooling. In practice, a well-defined scope fits in a sprint starting at 400 euros (excl. VAT) and a usable business application reaches production in 4 to 6 weeks.
Can you skip the developer entirely thanks to AI?
For a throwaway prototype, sometimes. For a tool that holds customer data in production, no: Google's 2025 DORA report shows 30% of professionals have little or no trust in AI-generated code, and delivery instability remains a concern. Review, security and architecture decisions remain human work.
How do I know whether an off-the-shelf tool is enough instead of custom?
If your process is standard (invoicing, payroll, classic support), an off-the-shelf product is almost always the right answer, and AI changes nothing about that. Custom software is justified when the process in question is where your value lives and no existing tool fits it without contortions. A 30-minute scoping call is usually enough to settle it.