Build vs Buy Internal Tools With AI: Which Should You Choose?

Should you build internal tools with AI instead of buying SaaS?

Yes. Build custom AI software now for the parts of your stack that are workflow automation, internal admin, and reporting. You don't need to wait for AGI, and you don't need to wait for the industry's safety debate to resolve. That debate is about something else entirely.

Why does the frontier debate have nothing to do with your internal tools?

In September 2026, Anthropic CEO Dario Amodei published "We Must Pace the Frontier," arguing the AI industry needs to deliberately slow the rate at which it improves model capabilities. His concern wasn't that current models are too weak to be useful. It was the opposite: AI has started building the next generation of AI, and capability is compounding faster than the field's ability to understand and steer it. Sam Altman posted his agreement. Elon Musk replied with three words: "Dario is right."

That's a real argument, about recursive self-improvement, existential risk, and the pace of frontier research at the very top of the industry. It isn't a claim that GPT-6-class or Claude-class models today can't write a working internal dashboard or automate a reporting workflow. Conflating the two is the trap. Read "we must pace the frontier" and conclude "so we should wait to build," and you're answering a question nobody asked.

Can today's AI models already build your internal tools?

Set the AGI debate aside and look at what current models can do unsupervised. METR, the group that benchmarks AI agent capability on real tasks, tracks how long a coding task an agent can complete autonomously at 50% reliability. That time horizon has doubled roughly every seven months since 2019. Current frontier agents reliably finish tasks that take a skilled human close to an hour, and are starting to chip away at multi-hour tasks, even as reliability drops the longer the clock runs.

An hour-plus of unsupervised, reliable task completion isn't a research curiosity. It's roughly the shape of the work internal tools exist to do: pull data from three systems, reconcile it, render a dashboard, gate an approval, send a notification. That's the category SaaS was built to serve for teams without spare engineering time. It's also the category current models already handle well, a point covered in prior reporting on frontend generation eating niche SaaS.

Did Wall Street already flinch?

You don't have to take a blog's word that this is disruptive. In early February 2026, software and services stocks lost roughly $830 billion in market value over six trading days, triggered by a new Claude agent tool that pushed into the application layer traditionally owned by SaaS vendors. Nvidia's Jensen Huang called the idea that AI would replace software layers "illogical." JPMorgan's Mark Murphy pushed back on the notion that a plug-in from an LLM replaces "every layer of mission-critical enterprise software."

Both things can be true at once: the disruption is real enough to move $830 billion in a week, and it isn't evenly distributed across every SaaS category. That contested read is exactly why practical builders shouldn't wait for a clean consensus before they act. The market won't hand you certainty. It's already pricing in the uncertainty.

Why aren't builders waiting for permission?

While executives argue about pace, the people running workflows have moved on. Retool's 2026 Build vs Buy Report, surveying 817 builders, found that 35% of teams have already replaced at least one SaaS tool with a custom AI-assisted build, and 78% expect to build more in 2026. Sixty percent of builders across seniority levels have shipped something outside IT oversight in the past year. ClickUp built six internal AI tools and cut $200,000 a year in automation software spend doing it.

This isn't confined to a few aggressive teams. PagerDuty's shadow AI survey found 66% of office professionals admit to using AI tools at work they believed were against company policy, and 88% have shared work-related information with public tools like ChatGPT, Claude, or Gemini. Gartner, looking at the same trend from the vendor side, forecasts up to $234 billion in enterprise application spend at risk from "agentic arbitrage," where agents complete cross-system tasks and bypass the SaaS interface entirely, through 2030. Gartner analyst George Brocklehurst called it "less an apocalypse and more of a metamorphosis."

The throughline across Retool, PagerDuty, and Gartner is the same: employees and small teams aren't waiting on a research roadmap or a governance policy to decide whether current models are good enough. They've already decided. Anyone still treating this as a future-tense question, as covered in how build vs buy internal tools with AI became the default rather than the debate, is behind the people they manage.

When should you still buy instead of build?

None of this means every SaaS renewal is dead. Gartner separately predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls, and notes that of the thousands of agentic AI vendors on the market, only around 130 are judged real. That's a caution against sloppy scoping, not a case for standing still.

Analyst Alex Warfel makes the sharper version of the same point: AI genuinely lowers the bar for simple internal dashboards, lightweight workflow automation, and one-off data tools, but for software that is compliance-heavy, integration-dense, or mission-critical, the moat was never the code itself. Payroll, ERP, and regulated compliance systems carry integration debt, audit requirements, and liability that a fast first build doesn't erase. Gartner's own customer-service research backs this up in a different domain: it expects half of companies that cut support headcount because of AI to rehire by 2027, because AI still lacks the judgment and empathy the work requires. Cheap code isn't the same as low risk.

What does waiting to build actually cost you?

Whatever side of the AGI-pacing debate you land on, your SaaS bill isn't pausing to let you think it over. Gartner forecasts worldwide business software spending will grow 14.7% in 2026, topping $1.4 trillion, up from 11.5% growth in 2025: more than $180 billion in net new spend in a single year. That growth is happening independent of whether Amodei, Altman, and Musk agree on how fast frontier labs should move. The financial pressure to act is a today problem, not a wait-for-AGI problem, and it's the same pressure behind the broader economics of what a growing subscription stack actually costs.

What should you build now, and what should you keep buying?

Use the categories the data actually supports, not a blanket rule in either direction.

• Workflow automation. Retool's survey flags this as the single most at-risk SaaS category, at 35% already replaced or in progress. If the tool's job is moving data between systems and triggering actions, current models handle it well.
• Internal admin tools. Approval flows, request trackers, and internal CRUD apps sit close behind, at 33%. These rarely touch regulated data and rarely need vendor-grade uptime guarantees.
• BI and reporting dashboards. At 29% replacement, this is the category where a prompt now produces in an afternoon what used to be a procurement ticket and a monthly invoice.
• Compliance, payroll, and ERP. Keep buying here for now. The integration surface, audit trail, and liability exposure haven't gotten cheaper just because the code did.
• Anything a shadow build already exists for. If 60% of builders are shipping tools outside IT oversight, bring that work into the light instead of pretending it isn't happening. Give teams a sanctioned path, whether that's an internal platform team or an approved AI builder tool, instead of a ban nobody follows.

Run each candidate through three questions before you commit: does it touch regulated data, does it need to integrate with more than two or three external systems, and would a failure be expensive enough to need a vendor's SLA behind it. Two or more yes answers, keep buying. Mostly no, you have your next build. The practitioner's process for turning that decision into a shipped tool in an afternoon is more mechanical than most teams expect.

Should you pace the frontier of AI research or your own roadmap?

Amodei's argument for slowing frontier capability research is serious, aimed at a handful of labs racing toward systems that can improve themselves faster than anyone can audit them. It has nothing useful to say about whether a 200-person company should keep paying $50,000 a year for a workflow tool a current model could rebuild this quarter. Those are different questions running on different clocks, and caution on one doesn't license caution on the other.

The evidence is already in from the people closest to the work: over a third of builders have replaced a SaaS tool already, most plan to build more, and employees are quietly doing this with or without a policy in place. Software spending keeps climbing while you deliberate. Pick the tools on your renewal list that are workflow automation, internal admin, or reporting, run them through the three questions above, and start building the ones that clear the bar. Leave the AGI debate to the labs actually racing toward it.