Harvey vs. Legora: Inside the $16.6B Legal AI War That Will Restructure Professional Services

On February 17, 2026, Retool published the findings of a survey of 817 enterprise builders — engineers, operations leads, product managers, and IT administrators at companies ranging from well-funded startups to Fortune 500 corporations. The headline number should alarm every SaaS vendor in the market: 35% of respondents had already replaced the functionality of at least one commercial SaaS tool with a custom internal software build. Another 78% said they plan to build more custom internal tools in 2026.

The report's title captures the structural shift plainly: "The Build vs. Buy Shift: How Vibe Coding and Shadow IT Have Reshaped Enterprise Software." Two years ago, that title described a hypothetical. Today it describes an ongoing structural reordering of enterprise software procurement — one driven not by dissatisfaction with SaaS as a delivery model, but by a fundamental change in the economic calculus that has governed build-versus-buy decisions for 25 years.

The build versus buy decision is as old as enterprise software itself. What changed in 2024 and 2025 is the math underlying that decision. AI-assisted development — AI code editors, vibe coding platforms, and low-code tools supercharged by large language models — has collapsed the cost and time required to build custom internal software. Prototyping that used to take two months of senior engineering time now takes two days of operations management time. Deployments that used to require specialized infrastructure teams can be handled by a single developer in a week. The total cost of a custom internal tool has dropped by an estimated 10x in three years. Commercial SaaS pricing has not adjusted.

The Numbers Behind the Revolt

Retool's 2026 Build vs. Buy Report documents the revolt across five critical metrics.

Replacement is not hypothetical. 35% of respondents have already replaced at least one SaaS tool with a custom build. This is completed action, not stated intent. The decision to stop paying for a SaaS subscription and build something internally has already been made at more than a third of the companies surveyed.

The pipeline is accelerating. 78% plan to build more custom internal tools in 2026. Organizations that have done it once are doing it again. The cost and expertise barriers that previously made custom builds feel risky have been lowered by the same AI tools that made the first build succeed.

Shadow IT is endemic. 60% of respondents built software outside of official IT oversight in the past 12 months. 25% report doing so frequently. Enterprise governance has not kept pace with build capability.

AI-assisted productivity is measurable. 51% of respondents have built production software currently in use by their teams using AI assistance. Approximately half of those report saving six or more hours per week per person on tasks the custom software now handles.

The economics are closing fast. According to the VentureBeat analysis of the broader enterprise software shift, AI-assisted development has lowered custom software build costs to the point where the break-even period against SaaS subscriptions now measures in months rather than years for most mid-market use cases.

Which SaaS Categories Face Existential Pressure

The build revolt is not evenly distributed across software categories. Retool's data identifies a clear ranking of SaaS segments where the replacement rate is highest:

The pattern across these categories is consistent. The SaaS tools under greatest pressure are those where value is delivered through aggregation and convenience rather than through defensible architectural complexity. A workflow automation platform that connects common SaaS APIs does not require years of engineering to replicate — it requires a few days with an AI code editor and familiarity with the internal systems that need to be connected. When the build cost of that replacement has dropped by 10x, the subscription math no longer favors buying.

Workflow automation and internal admin tools lead the list because these categories are inherently organization-specific. No two companies share the same internal data flows, approval chains, or legacy system integrations. A generic automation platform optimized for the median customer is frequently the wrong shape for any specific enterprise. When that mismatch costs $40,000 per year in a subscription and $12,000 to build something purpose-fit, operations leads do the math themselves.

The Economics That Changed the Calculus

The build versus buy calculation has shifted across three reinforcing dimensions simultaneously, making the current inflection different from previous waves of shadow IT or internal tooling adoption.

Build cost has collapsed. The standard engineering estimate for a custom internal tool two years ago was $50,000 to $150,000 in initial development cost, plus ongoing maintenance. AI-assisted development has compressed this to $5,000 to $20,000 for comparable functionality, with maintenance costs reduced by AI-assisted debugging and refactoring. The prototyping phase alone has compressed by 70 to 80% for the categories most exposed to replacement. This moves the break-even point against SaaS subscriptions from 2-3 years to 6-12 months.

SaaS pricing has continued to increase. Average enterprise SaaS spend rose 8% year-over-year to $55.7 million annually per organization in 2026, according to BetterCloud's annual market survey. Application portfolios have remained essentially flat at around 305 applications per organization — all the cost increase is coming from price inflation, AI-tier upsells, consumption surcharges, and contract expansion pressure. Organizations are paying more for SaaS at precisely the moment that building alternatives has become dramatically cheaper. The economic pressure from both directions is converging.

Operational complexity has declined. Hosting, deployment, and maintenance of custom software were historically meaningful ongoing costs that made the total cost of custom builds substantially higher than initial development estimates. Cloud-native deployment tooling, AI-assisted code maintenance, and the commoditization of infrastructure services have reduced ongoing operational overhead for custom builds significantly. The hidden costs that used to tilt build-versus-buy calculations back toward buying have largely been eliminated for the categories under greatest replacement pressure.

The Shadow IT Crisis That Nobody Is Managing

The 60% shadow IT finding in Retool's report deserves more attention than it typically receives in enterprise software discussion. Shadow IT — software built or deployed outside official IT governance — is not a new phenomenon. What is categorically new is the sophistication and operational criticality of what is being built.

Previous waves of shadow IT involved consumer tools repurposed for business use: employees storing corporate files in personal cloud accounts, building spreadsheets that accumulated into business-critical data infrastructure, or subscribing to marketing SaaS tools through personal credit cards. These were visible, auditable, and bounded — they could not easily create new operational dependencies that were invisible to the IT organization.

The 2026 wave is different because the outputs are custom software running against internal systems. AI-assisted development tools have lowered the barrier to building production software to the point where operations managers, marketing analysts, and finance professionals are writing and deploying Python scripts, Retool applications, and API integrations that process internal business data and run operational workflows outside of any security review, access control governance, or documentation standard.

This is not inherently bad — the custom builds represent genuine value creation, often addressing real workflow gaps that commercial SaaS products couldn't fill at reasonable cost. But enterprise governance frameworks were built for a world where building production software required specialized engineering skills and approved tooling chains. That world no longer exists, and most enterprises have not updated their governance accordingly.

The audit and operational risk from this gap is concrete. A custom billing integration built by a finance analyst running on a personal AWS account, accessing production payment data, without documentation or backup: when that analyst leaves, what happens to billing operations? An AI-assisted data pipeline processing customer information outside of data governance policies: how does the organization respond to a GDPR data subject request that implicates a system it does not know exists?

What This Means for SaaS Vendors and Their Moats

The build revolt does not spell the end of commercial SaaS. It spells the end of undifferentiated SaaS priced as though building alternatives were still expensive.

The SaaS products that will survive this pressure share specific characteristics. They provide value through genuine architectural complexity that cannot be replicated in days or weeks. They encode years of enterprise process, regulatory compliance infrastructure, or ecosystem relationships that represent real switching costs independent of switching from subscription to build. Cursor's remarkable $2B ARR trajectory demonstrates that AI-native developer tools — the category enabling the build revolt — command strong willingness-to-pay precisely because they are not easily built internally.

Salesforce's core CRM product is not primarily at risk because its value is not the form fields and workflow automation — it is 25 years of enterprise relationship management process encoded in software, integrated with thousands of third-party systems, supported by an ecosystem of ISVs and certified consultants, and embedded in procurement and legal processes that resist displacement. That is not replaceable by an operations lead with a Retool account.

The products under genuine existential pressure from the build revolt are those where value delivery is primarily through API aggregation and convenience rather than defensible architectural complexity. Workflow automation tools that connect common SaaS APIs. BI dashboards that sit on top of standard data warehouse connectors. Internal admin tools with limited customization. Project management platforms with generic templates. These categories will face sustained replacement pressure from organizations that have now learned they can build better-fit solutions for a fraction of the subscription cost.

The PLG and enterprise SaaS growth models are both being stressed by this shift in complementary ways. PLG products relying on high-conversion free-tier funnels are seeing enterprise conversion rates decline as engineering teams build custom alternatives before reaching enterprise pricing thresholds. Enterprise SaaS products relying on organizational inertia and migration cost moats are finding those moats lowered by AI-assisted development that reduces migration effort substantially.

How to Think About Build vs. Buy in 2026

For product teams evaluating commercial SaaS versus custom builds, the decision framework needs to reflect the new economics rather than the assumptions from three years ago.

1. Re-estimate build cost using AI-assisted development rates. The default engineering estimates most organizations use are based on pre-AI tooling assumptions. A realistic current cost for a custom internal tool should assume a skilled developer with AI coding assistance, not a team of senior engineers. A tool that would have cost $100,000 to build in 2022 likely costs $12,000 to $25,000 today. Adjust the break-even calculation accordingly before choosing SaaS.

2. Calculate the customization tax explicitly. Before signing a SaaS contract, document how much of the vendor's standard feature set you will actually use, and what the full cost of customization — professional services, admin overhead, and ongoing configuration maintenance — will be for your specific workflow. If the customization cost exceeds 35% of the first-year contract value, the custom build case is materially stronger than a surface-level comparison suggests.

3. Build governance infrastructure before you build software. The shadow IT crisis documented in Retool's report is primarily a governance failure, not a technology failure. Organizations that want to capture the value of custom builds without creating operational and compliance risk need to establish clear policies for internal tool development: what data can be accessed, what deployment standards apply, what documentation is required. This infrastructure takes weeks to build and prevents years of compounding technical and regulatory risk.

4. Treat build decisions as portfolio management. The organizations succeeding with the build revolt are not building everything indiscriminately — they are making deliberate portfolio decisions about which categories justify building and which are better served by commercial SaaS. The relevant question is not "should we build or buy?" but "in which specific category, for which specific workflow, does building generate better long-term ROI than purchasing?" The answer differs by category, workflow complexity, and internal build capability.

What the Vibe Coding Wave Reveals About the Long-Term Cost

The irony embedded in the build revolt narrative is that the tools enabling enterprises to replace SaaS products are themselves SaaS products. Cursor, GitHub Copilot, Codeium, and similar AI coding tools are subscription software that enterprises pay for — and use to build replacements for other subscription software.

This points to a bifurcation forming in the SaaS market. AI-native developer tools that empower their users to reduce dependencies on other SaaS categories are demonstrating strong retention and willingness-to-pay precisely because the value they deliver is measurable and compounding. The vibe coding and shadow IT wave is built on these tools.

The longer-term risk for custom builds built quickly with AI assistance is technical debt. Research on AI-generated code has consistently found that code written with AI assistance tends to be functionally correct on first deployment while introducing subtle architectural patterns that compound into maintenance burden over 12 to 24 months. Organizations that are building their way out of SaaS subscriptions need to monitor their custom tool portfolio for this risk — not just at initial deployment, but as the tools age and the developers who built them move to other projects.

The build revolt is not the end of enterprise SaaS. It is a forcing function that will ruthlessly separate SaaS products with genuine moats from those that have been selling convenience at enterprise pricing.

Takeaway: The enterprise build revolt is real, data-documented, and accelerating. Retool's 2026 Build vs. Buy Report records that 35% of enterprise teams have already replaced at least one SaaS tool with a custom AI-assisted build, and 78% plan to build more. The economics driving this shift — AI-assisted development reducing build costs by 10x while SaaS pricing continues to increase — are structural, not cyclical. SaaS vendors without defensible architectural moats, regulatory complexity, or deep ecosystem integration are exposed. Product teams navigating this environment need to update their build versus buy frameworks to reflect 2026 economics, build internal governance infrastructure to manage shadow IT risk, and make portfolio-level decisions about where custom builds generate sustainable ROI and where commercial SaaS remains the more durable choice.