In February, a Claude update triggered a massive selloff that seemed to signal the “SaaSpocalypse” was nigh. But the market really just caught up to what builders already knew—that companies of any size are now capable of building the tools that allow them to move faster or invest more into their competitive edge.
Building custom used to be reserved for the companies that had the time, headcount, and cashflow to afford it. Now, as LLMs have improved and vibe coding’s gone mainstream, the traditional “build vs. buy” economics have shifted. It’s harder to justify spending thousands on a subscription to a generic tool—with additional costs for customization and upgrades—when your team could have a prototype tailor-made to solve your problem in a few days.
But if AI’s impact on the labor market is any indication, AI replacement (of humans or software) isn’t guaranteed. Companies are already expected to rehire for roles they had thought to replace with AI. It’s a cautionary tale: good AI alone isn’t enough to solve a business problem. Connectivity, security, and governance will determine which custom builds drive transformational change at the company level.
We surveyed 817 Retool customers and builders and found that 35% of them have already replaced at least one SaaS tool with a custom build, and 78% expect to build more of their own tools in 2026. Businesses that want their tooling to be a competitive advantage need to build for the long-term with the right solutions. Enterprise AppGen platforms like Retool, where builders can generate apps on their data using natural language, will help organizations build custom tools securely to give them a competitive edge.
Just over one-third of respondents (36%) identified as software engineers or developers. The rest spanned operations (12%), product management (10%), data (7%), marketing and sales ops (6%), IT (6%), business analysis (4%), finance (4%), and other roles (16%).
- Shadow IT is no longer an edge case. 60% of builders across levels of seniority have built something outside IT oversight in the past year.
- AI mandates keep growing, but measurement hasn’t caught up. 75% of builders now work under AI directives, yet 35% of organizations still haven’t established any AI productivity metrics.
- Most builders are shipping real software with AI. 51% have built production software currently in use by their teams, and about half of those report saving six or more hours per week.
- Vibe coding at work looks different than on the weekend. Among those who’ve shipped software, 72% use AI to write discrete pieces of code they integrate into larger projects—only 31% are prompting their way to complete apps.
- Confidence in AI-generated code is growing but cautious. Only 8% use AI code without changes. Just under half of builders (44%) test thoroughly before deploying anything.
When a business user can build a custom solution in a weekend that does exactly what they need, “good enough” generic software stops being, well, good enough.
35% of teams have already replaced functionality of at least one SaaS tool, and 78% plan to build more custom tools in 2026.
If this trend persists without proper governance, the SaaS landscape could see an explosion of low-cost, one-off tools that prioritize speed-to-build over quality (what Sam Altman called the “fast fashion era” of SaaS). The future of software lies somewhere in between stodgy, inflexible SaaS and sloppy, ungoverned sprawl, where builders can solve their problems within secure, governed environments.
The survey found that every SaaS category is under replacement pressure. Workflow automations (35%) and internal admin tools (33%) top the list—areas where the gap between what a SaaS tool provides and what an organization actually needs is often widest.
BI tools (29%), CRMs and form builders (25%), project management (23%), and customer support (21%) are all under pressure, too. It’s clear the competitive bar for purchased software has just gotten much higher.
While anyone can replace a SaaS tool with AI, most vibe-coded apps only replicate a feature because it has no idea how your business actually runs: what data lives where, who should access it, or how it stays compliant. Enterprise AppGen makes those custom replacements durable.
ClickUp, a productivity platform serving 14 million users, evaluated a wave of AI vendors for its GTM operations. None had the right integrations or staying power.
“We realized we could build these tools ourselves and save on multiple subscriptions,” says Borys Aptekar, GTM AI Product Manager.
Instead, they built six AI tools connected to Salesforce, Zendesk, and Snowflake that automated hundreds of hours of weekly work, saved hundreds of thousands in headcount costs, and cut $200K per year in automation software.
Harmonic, a startup discovery platform, tells a similar story. Miles Konstantin, Head of Automation and Tooling, hit a breaking point with a $20,000-per-year third-party tool. “Their support was so slow that it was faster for me to rebuild the product inside Retool than wait for support to get back to me,” he says.
That rebuild sparked a cultural shift toward building. Harmonic now runs 33 internal apps connected to Salesforce, Gong, Slack, and internal APIs with audit logs and role-based access built in. When someone wants new software, the default question is now: “Why can’t we just build this in Retool?”
When a custom tool saves teams money and time, connects to live data, and ships with enterprise-grade security from day one, it becomes harder to justify the “buy.”
Shadow IT—software, hardware, or other workplace services unauthorized by IT—is spreading. Builders are solving their own problems with custom tools faster than their organizations can procure and approve something new.
We found that 60% of builders have built something outside of IT oversight in the past year, and 25% report doing so frequently.
Organizations can choose to channel this energy into something real and productive. Shadow IT is a signal: the traditional methods of SaaS procurement are outdated, and high-value unauthorized tools might have a place in a secure, managed environment.
When we asked these builders why they went outside official channels, 31% cited speed, 25% cited unmet needs, and 18% said IT’s process was too slow. These aren’t rogue actors—64% of our respondent pool were senior managers and above. They’re experienced builders choosing to move fast over moving officially.
How those tools connect to each other, and to a business’s internal systems, matters. Builders going around IT to replace 10 disconnected spreadsheets with 10 disconnected pieces of custom software won’t enable faster work or company-level transformation.
When we asked the builders working in the shadows about what they built, we found the categories that dominate—internal tools (53%), automated workflows (53%), custom dashboards (51%)—are the kinds of traditional SaaS tools that require engineering resources and IT approval.
Tools that save time are another incentive to build in the shadows. Among respondents who’ve built production software in use by their teams, about half (49%) say their tool saves six or more hours per week.
The time savings are real and substantial, but the risks are, too. Time saved on one task can be offset by time lost to maintenance, wrangling ungoverned tool sprawl, or triaging security vulnerabilities if not built in the right environment. Organizations need to prepare with the right systems and guardrails in place: corporate policies, centralized AI resources, and solutions built for the moment.
Vibe coding—prompting your way to a prototype—is great for personal projects and POCs. It’s also transformed how builders start projects with an efficient path to a prototype. Getting to deployment, however, requires more than a few prompts.
We found that about half of builders (51%) are, in fact, shipping production software with AI, and another 17% have at least experimented.
18% haven’t tried to build anything at work yet, and 14% have tried, but their software hasn’t reached production.
“There’s no way you can go live with a vibe-coded solution,” says Pierre Yves Calloc’h, speaking about his experience building at Pernod Ricard. “It might work for demos, but we build enterprise-grade technology that has to scale across 30 countries.”
When it comes to building with AI, the most visible, viral tools are often the ones used for personal side projects: the result of ambitious builders tinkering and testing boundaries. We wanted to know how vibecoded tools are being built at work, and we specified that in our questions.
LLMs and coding assistants dominated among tools builders use at work, and AI-only app builders lag when it comes to work software. Just 31% of builders report prompting their way into an app at work, so it follows that they’re leaning on a few tools at a time that support different pieces of the development process.
Almost all builders (93%) are using LLMs to code, build, or automate at work. ChatGPT leads among LLMs at 70%, but the distribution tells us builders aren’t loyal to a single model. Gemini and Claude both exceed 50% adoption, suggesting a multi-model approach has become standard practice.
The model race is also constantly shifting, with Claude Code’s explosion in late 2025 and the launch of Cowork giving it an edge in the hype cycle. Gemini’s model improvements in 2025 also drove Sam Altman to declare a “code red” last year. The landscape is always evolving, and builders will evolve with it.
Most builders are also using coding assistants at work—AI tools that can write and debug code. GitHub Copilot (39%) and Cursor (35%) emerged as the clear leaders, but code alone can only get you so far—28% of respondents don’t use AI coding assistants at work.
AI-only app builders—tools like Lovable, v0, and Replit that emphasize prompts and minimize or eliminate the need for code—are now some of traditional SaaS’s most visible challengers. They’re great for prototypes or weekend projects, but production software needs easy access to code and the ability to connect to enterprise data securely.
The AI-only app-building category is nascent and fragmented, and despite its virality, more than half of respondents (56%) don’t use AI-only app builders at work.
Most AI-only app builders optimize for speed to prototype, not speed to production. These AI-only builders also own the infrastructure. Enterprise customers trying to replace their clunky SaaS tools will get a quick prototype, but without control over security, permissions, and where the data lives.
Enterprise AppGen platforms like Retool, by contrast, allow business users to prompt with AI while keeping their orgs’ data in their systems. and the apps inherit their existing security model. They provide a clear, sanctioned path to production with guardrails that enterprises need.
There are two ways to build software with AI right now: use natural language to prompt your way to an app, or write the code yourself with help from AI. One gives you speed, the other gives you control. Production-grade custom software needs a platform like enterprise AppGen that supports both and governs the result.
The popularity of LLMs and coding assistants among builders we surveyed tracks with how they’re actually building. Most builders who’ve shipped working software (72%) are using AI to write discrete pieces of code, which they then integrate into larger projects.
57% use AI to debug, fix, and refine their code, and 48% use it for planning or outlining their builds. Only 31% of builders are prompting their way to complete applications.
The 72% writing code snippets are trading speed for control. They can (and do) test and verify each piece before it touches production. The 31% prompting apps are trading control for speed. They get something functional fast, but with less visibility into what the AI actually built. That tradeoff matters most when builders are replacing existing software. A code snippet you can test and integrate is safer than a prompted app you can’t fully inspect, but it’s also slower and requires more technical skill. The teams replacing SaaS tools at scale need both modes available, with a shared security and data layer underneath.
Speaking of testing code, about two-thirds of builders (63%) are at least somewhat confident in the code AI generates for them. But that confidence doesn’t mean builders are deploying unchecked.
Only 14% are very confident, while 20% are still not very confident in their AI-generated code. This caution shows up in their workflows. Only 8% of these builders use code immediately without any changes.
Most builders aren’t copy-pasting AI generated code—44% of them test their AI-generated code thoroughly before using it, and 32% at least review code briefly.
Code reviews have been the standard on most engineering teams for a long time. What’s changed is the volume: more people can write code, which means more code needs reviewing. Organizations scaling AI-powered building will need to scale their review processes alongside it.
Half of builders have shipped production software with AI. The other half are stuck hitting walls that speed or code alone can’t solve.
The blockers split roughly into two categories: capability gaps (technical knowledge, hallucinated code, missing secure data integrations) and organizational friction (lost priority, insufficient time, IT approval issues).
No single blocker stood out, so it’s clear a new builder could run into any or all of these hurdles. The governance gap isn’t showing up as a deployment blocker, it’s showing up as shadow IT.
Only 17% said they were blocked by building something unsanctioned by IT, but that number is deceptively low. Remember: 60% of builders have already built something outside IT oversight entirely. IT can’t block SaaS builds they don’t know about. And when those ungoverned tools connect to production data or replace existing SaaS workflows, the risk compounds quietly until something breaks.
Still, builders are optimistic these blockers are temporary. About one-third of these respondents (32%) are still planning to deploy their apps. How much time these builders will lose in the process, though, might offset any potential time savings from those tools.
Right now, companies are hearing the siren song of rip-and-replace. But we have a recent AI cautionary tale playing out in parallel: companies that rushed to replace workers last year potentially rehiring them as soon as 2027. SaaS replacement could fall into the same trap if businesses don’t invest in the right tools and ways to measure the outcomes.
The lesson for software is the same as the lesson for headcount: replacement without measurement is just churn. And right now, most organizations aren’t measuring much at all.
Almost all builders (91%) reported their company had reached some level of basic AI automation maturity.
Still, it may be a while before we see widespread workforce automation. Nearly three-quarters of organizations (72%) are still at Basic or Intermediate stages. Only 19% describe themselves as Advanced.
More custom tools doesn’t equal automation maturity. The maturity gap comes down to who is replacing the right things. Teams nearing advanced maturity are automating where traditional SaaS tools can’t keep up with how they actually work.
But most organizations aren’t there yet. When we asked what's holding them back, the blockers split into two categories: the technical and the organizational.
We asked about both technical and organizational barriers to automating more work with AI.
The top technical blockers:
- Lack of tech resources/engineering bandwidth (42%)
- Security and compliance concerns (41%)
- Integration challenges between systems (39%)
The top organizational blockers:
- Unclear ROI on the project (33%)
- Budget constraints (30%)
- Maintenance burdens (26%)
Security and compliance concerns are pronounced on the technical side—a reflection of real risks that organizations are right to take seriously. When builders create AI-powered tools that connect to production data, the security model of those tools matters enormously.
The 2025 Builder Report found that enterprise AppGen platforms create the right conditions for scalable productivity by bridging the gap between AI code generation and production deployment. Without that bridge, security concerns become deployment blockers.
But organizational barriers might be even more significant when it comes to company-level adoption. Unclear ROI sits at the top of the list, and you can’t prove returns on what you don’t measure.
Leaders know AI should be making their organizations more productive, and builders are clearly eager to ship software as unique as their problems. But you can’t prove ROI on what you don’t measure, and 35% of organizations still haven’t established any AI productivity metrics.
AI is moving unbelievably fast, but it’s still new in the grand scheme of software development. You have to experiment to measure. But experiment too long, and it’ll be harder to tell which experiments are really driving ROI.
For organizations to truly transform, they need to move beyond anecdata and experimentation. Our 2025 Builder Report found that 74% of builders under AI mandates reported exceeding leadership’s expectations, but only where organizations had visibility into what builders were actually producing.
Measurement kickstarts feedback loops that allow organizations to double down on what’s working. Traditional SaaS subscriptions had line items—a clear monetary value that made questions about ROI more straightforward to answer. Custom-built tools that save teams time but exist on the IT fringes get harder to justify, and it’ll be easier to feel the pain of ungoverned sprawl as it grows unchecked. Organizations building time-saving tools in a controlled environment will know what’s working, cut what isn’t, and prove the ROI that justifies building over buying in the first place.
Builders have already figured out they can replace software—35% of teams have already replaced at least one purchased tool with something custom-built, and 78% expect to build more in 2026. Custom software can give organizations a competitive edge, but if hundreds of unsanctioned point solutions are allowed to proliferate across departments, SaaS sprawl becomes a liability.
With AI, all builders can ship fast, but sustaining what they ship requires infrastructure most organizations don’t have in place yet. Enterprise AppGen solves this. It connects AI prototypes to production data with security and governance built in from the start. Learn more or try it for yourself today.
This report is based on a comprehensive survey conducted in late 2025 across 817 builders, including Retool customers. The survey employed a mixed-methods approach, combining quantitative multiple-choice questions with qualitative open-ended responses to capture both measurable trends and nuanced insights.
Reader
