The Short Version

‍

The Decision Pattern Emerging Across Enterprise Teams

There is a conversation showing up in enterprise software decisions right now.

A company is looking at a software licensing option. The contract crosses $50K, sometimes $150K, sometimes $500K. It is a core one or two feature tool they want to run and maintain themselves. And someone at the table says: why are we paying for this when we can build it and own it?

That question used to end quickly. The cost of building was prohibitive. The team was too thin. The risk of owning a custom codebase felt too high relative to the predictability of a SaaS subscription. So companies bought, integrated, maintained the integration, watched the vendor raise prices at renewal, and bought again.

That calculus has changed.

The tools that made SaaS the default choice for fifteen years, cost certainty, speed to deploy, no engineering overhead have been disrupted by the same AI wave that SaaS vendors are now scrambling to absorb. Building is faster than it has ever been. The total cost of owning a SaaS stack has never been more visible. And the compliance environment is, for the first time, actively penalising enterprises that cannot demonstrate control over their AI-enabled workflows.

This is not an argument for building everything. Systems of record like Epic, SAP, Workday's HR core are rational purchases and that will not change. The argument is narrower and more specific: for the category of non-core, workflow-specific, competitively relevant tooling that represents the bulk of enterprise SaaS spend, the default answer in 2026 is build.

The pattern suggests that many CTOs are still evaluating these decisions using 2019 assumptions against 2026 market conditions. Here is what has actually changed, and why the shift is happening now.

What Changed Between the SaaS Era and the AI Build Era

Somewhere along the lines between the SaaS Era and AI build era things have taken a shift which is structural and here are the forces behind that shift:

Force 1: Why SaaS TCO Looks Different in 2026

What is happening

The average enterprise operates 106 SaaS applications, down from a peak of 130 in 2022. That consolidation sounds like progress while it is not.

Total SaaS spend has not fallen proportionally. Reducing app count has concentrated spend into fewer vendors while leaving the integration debt, the shelfware, and the fragile connective tissue behind.

‍

What changed in 2025/2026

The evaluation cost has inverted for workflow-specific tooling:

‍

This does not mean full production systems are cheap. It means the up-front investment required to find out whether building is the right call has collapsed. That changes when the conversation happens and that is what is showing up at $50K–$500K renewal tables.

The pain for CTOs

The integration layer compounds the problem. Each SaaS tool carries its own API, its own versioning schedule, its own deprecation cadence. A change in one vendor's API breaks workflows across several others simultaneously.

• More than one-third of breaches involved shadow data, data stored in unmanaged sources (IBM Cost of a Data Breach Report)
• $4.88M global average cost per breach, with shadow data incidents costing 16% above that average
• Shadow data breaches also took 26% longer to identify and contain

A CTO with dozens of SaaS tools in their stack, many operating outside formal IT governance, has a risk surface they cannot fully see or audit.

The build implication

The honest TCO of a SaaS tool includes: license cost, integration development, ongoing maintenance, security audit overhead, and the cost of accepting the vendor's pricing decisions over the contract term. For workflow-specific tools, that number is significantly higher than the subscription invoice.

The build-versus-buy comparison that procurement teams run is usually comparing against only the license line. When enterprises run the full calculation, the 3-year crossover point for custom software development has moved into the range of tools that were previously unquestioned renewals.

"The question is not what it costs to build versus buy. The question is what it costs to buy, integrate, maintain, and accept dependency on, versus build the specific thing the team actually needs."

Force 2: The Shift From Six-Figure Prototypes to Low-Threshold MVPs

What is happening

The cost to validate and get a workflow-specific tool to a production-capable MVP has fallen from six figures to low thousands. That is not the same as saying enterprise-grade software is cheap, a full production system with SSO, RBAC, audit logging, CI/CD, testing, monitoring, and compliance still requires real engineering investment. What has changed is the threshold at which building becomes worth evaluating seriously.

In 2022, getting to a working prototype required significant upfront engineering commitment before you could even test whether the build was the right call. That front-loaded cost made buying the safe default. That barrier has gone.

Vendor-reported benchmarks from Lovable CEO Anton Osika:

*These are vendor-reported anecdotes from a press interview, not independently audited case studies.

‍

What changed in 2025/2026

Andrej Karpathy coined the term "vibe coding" on February 2, 2025. Collins Dictionary named it 2025 Word of the Year. More consequentially, the tools that year crossed a threshold: not faster development for engineers, but the removal of the engineering prerequisite entirely for a growing category of internal software.

• 75% of Replit's customers never write a single line of code (Replit CEO Amjad Masad, February 2025)
• 93% of enterprise builders now use LLMs to code, build, or automate at work (Retool 2026)
• Coding represented 55% of all departmental enterprise AI spend in 2025, $4 billion making it the single largest AI investment category (Menlo Ventures 2025 State of Generative AI)

None of this means the engineering discipline required for a production system has disappeared. What it means is that the cost and time to reach a point where that engineering work is clearly justified, a working prototype you can actually evaluate has collapsed.

What the tooling actually costs to get started

This is the cost to validate and reach a production-capable MVP for a workflow-specific internal tool, not an enterprise platform with full compliance, monitoring, and resilience built in. The point is that the entry cost for evaluating whether to build has collapsed, which changes the decision calculus at the point of SaaS renewal.

‍

A full enterprise deployment with SSO, RBAC, audit logging, security review, CI/CD, and observability requires additional engineering investment on top of this. The shift is in the cost and time to reach the point where that investment decision becomes well-informed rather than speculative.

The pain for CTOs

The objection that ended most internal build conversations for fifteen years was "we don't have the bandwidth" is expiring in real time. The team that was too small to build in 2022 has materially different throughput in 2026. The constraint has shifted from capacity to prioritisation.

The build implication

Every SaaS renewal conversation now has a credible challenger. For workflow automation, internal tooling, reporting layers, customer-facing portals, and integration middleware, the question "what would it cost to build this?" frequently comes in below the annual renewal.

Cursor reached $2 billion in ARR in approximately three years, confirmed as the fastest B2B ARR scaling on record, ahead of Slack, Zoom, and Snowflake. Lovable reportedly reached a $200M annualised revenue run rate within its first year. These are AI coding tools, and their adoption curves are the market's signal that the build threshold has moved. Enterprise procurement assumptions have not caught up to what those numbers represent.

The build economics behind these numbers apply directly to the custom AI agent development decisions enterprises are making right now, and to the forward deployed engineering model that makes those builds execute without the delivery risk that historically justified buying instead.

Force 3: Why Per-Seat Pricing Breaks in Agent-Driven Workflows

What is happening

The economics of SaaS were built on one assumption: each software user needs a license. That assumption is structurally broken in an agent-first enterprise. AI agents do not have seats.

When an enterprise's agents interact with a CRM 10,000 times per day on behalf of three human users, the vendor's three-seat pricing model has no relationship to actual value consumed.

Two of the most direct quotes on where this is heading:

Satya Nadella, BG2 Podcast, late 2024: "Business applications are essentially CRUD databases with a bunch of business logic. All that business logic is moving to AI agents."

Aaron Levie, Box CEO, March 2025: "AI Agents will be the biggest shift to enterprise software business models that we've ever seen. there may be little to no connection between the number of users on a platform and the total amount of usage of AI Agents."

What changed in 2025/2026

Enterprise AI agent deployment crossed from pilot programs to production deployment during 2025:

• Agent deployment tripled since Q4 2024 (KPMG AI Quarterly Pulse, June 2025)
• 82% of business leaders believe agents will become valuable contributors within a year
• Salesforce Agentforce crossed $500M ARR in Q3 FY2026, up 330% year-over-year

The software contracts CTOs signed in 2021–2023 were priced on the assumption that humans would operate the software. That assumption is now operationally false for a growing share of workflows.

The pain for CTOs

Vendors are repricing toward outcomes because per-seat models cannot survive agent-first usage patterns. For enterprises, that repricing is a permanent and escalating cost exposure against workflows they do not own. Every quarter on an outcome-based vendor contract is a quarter of paying for value that could be compounding inside the organisation instead.

The build implication

The logic: wherever AI agents can now execute workflows previously done by human teams or SaaS tools, the unit of competitive value is the workflow itself and not the license.

‍

Understanding agentic AI versus generative AI matters here, the distinction shapes which workflows are build candidates and which still require a vendor's specialised infrastructure.

If agents are now touching those workflows, the economics of that contract have already changed. The question is whether your renewal conversation will reflect that.

A Custom Software Scoping Session with Ideas2IT produces a build-versus-buy recommendation for a specific workflow within two weeks: full scope, architecture, delivery model, and 3-year TCO comparison run against the real cost of renewing. It is designed for teams that have already asked the build question and need a defensible answer before the next renewal table.

Start the Scoping Conversation →

‍

Force 4: The Regulatory Pressure Behind Enterprise Software Ownership

What is happening

The EU AI Act's high-risk AI system obligations become enforceable on August 2, 2026. They require high-risk AI systems to maintain auditable data lineage, explainability documentation, and human oversight mechanisms.

Note: The European Commission's Digital Omnibus proposal (February 2026) has suggested some procedural simplifications, but no blanket delays to the August 2 date have been accepted as of May 2026. Enterprises should treat August 2, 2026 as the operative deadline.

High-risk AI system categories under the Act:

• AI in underwriting decisions
• AI in hiring and HR screening
• AI in credit scoring
• AI in medical triage and clinical decision support
• AI in customer eligibility determination

Most large enterprises have already deployed AI in at least one of these categories. The question is whether their current architecture typically AI embedded in a SaaS-hosted platform can provide the documentation a regulator will require.

"Our vendor is working on it" is not a defensible answer in a DPA audit.

What changed in 2025/2026

In 2022, compliance was a consideration for regulated industries. In 2026, it is a deadline with a penalty structure for any enterprise running AI in consequential workflows, regardless of industry.

‍

The pain for CTOs

Owning a custom build is one of the few architectural paths that gives a CTO direct, unmediated control over:

• Data residency and localisation
• Access logging and audit trails
• Model versioning and explainability documentation
• Human oversight mechanism design

Other options exist like dedicated tenancy, sovereign cloud, self-hosted enterprise platforms, but they depend on what vendors offer and on timelines they control. For enterprises running AI in high-risk decision workflows, the question is whether the vendor can provide the documentation a regulator requires, on the schedule the regulator demands. Ownership removes that dependency entirely.

The build implication

The regulatory environment has shifted the default risk posture from "buy and accept vendor terms" to "verify the vendor's compliance posture to an auditable standard or own the stack directly." For many enterprises in regulated industries, verification alone is proving insufficient. Vendors cannot always provide the documentation required on the schedule regulators demand, and the enterprise carries the liability either way.

Gartner's 2025 analysis identified data sovereignty as the number one driver of multicloud architecture investment in regulated industries. That is the budget signal: enterprises are already paying to own their data posture. For AI-enabled workflows in consequential decision categories, owning the software that processes that data is the most direct way to make the compliance answer defensible.

For teams working in healthcare specifically, this connects directly to why health systems are building AI capabilities beyond traditional EHR vendors rather than waiting for vendor roadmaps to catch up to compliance requirements. The same logic applies across financial services, insurance, and any enterprise deploying AI in consequential decisions.

Force 5: Why Competitive Advantage Now Lives in Proprietary Workflows

What is happening

For fifteen years, buying best-in-class SaaS was a reasonable competitive strategy. Every company bought the same tools, so advantage came from how well teams used them instead of which platforms they ran.

That logic held while SaaS products were genuinely differentiated by proprietary capabilities: ML models built over years, proprietary data sets, algorithms that took real engineering investment to produce.

AI has commoditised the functional layer of most workflow tools. When every CRM uses the same foundation models for lead scoring, the CRM's AI is not a competitive differentiator. It is a feature shipped to every customer on the same release cycle. Whatever differentiation it creates accretes to the industry, not to any individual enterprise that purchased it.

What changed in 2025/2026

The vendor's defensibility has shifted from AI capability to workflow integrations and data retention. That means the enterprise that owns the workflow and the data owns the competitive position and the vendor who owns them owns it instead.

The pain for CTOs

The ceiling on operational differentiation is set by the vendor's product roadmap. The vendor's incentive is to build for the broadest possible customer segment, which by definition dilutes any proprietary advantage the feature created for early adopters. Every improvement ships to every competitor on the same day.

A SaaS-dependent architecture is structurally limited in how much it can differentiate from competitors using the same stack.

The build implication

Retool's 2026 report documented the behavioural shift at the team level. Harmonic now runs 33 internal applications. Their default question at any new tool evaluation: "Why can't we just build this?"

That inversion from "why build" to "why not build" is the sign of an organisation that has absorbed the changed economics and is compounding on them.

The companies building competitive moat in 2026 are doing it through proprietary data, proprietary workflows, and domain-specific models that no vendor can replicate. These are not available on a pricing page. They are built. The enterprises building them now are accumulating an advantage that will be visible in 18 months.

"Every quarter spent on a vendor's roadmap is a quarter your competitors are compounding on their own."

This is why enterprise AI strategy execution is increasingly centred on build decisions rather than vendor selection decisions

What Makes a Workflow a Strong Build Candidate

Not every SaaS tool is a build candidate. The decision depends on the specific workflow, team context, and time horizon. This framework is a starting point, not a formula.

Build is the stronger choice when:

• The workflow is specific to how your business operates and unlikely to be served well by a generic vendor roadmap
• AI agents will be the primary software interface within 12 months
• Regulatory requirements demand auditable control over data residency, model decisions, or audit trails
• The 3-year TCO including integration, maintenance, and vendor pricing risk exceeds what a build would cost
• The workflow is where your competitive differentiation actually lives

Buy remains rational when:

• The tool is a system of record with deep compliance history (Epic, SAP, Workday HCM core)
• The vendor's specialisation is the product, the capability would take years and substantial investment to replicate
• The workflow is standard across your industry and differentiation comes from execution
• Your team does not have the governance model or delivery structure to own a custom build without it becoming technical debt

For a more specific framework in the context of medical software, the CTO's build vs. buy guide for custom medical imaging software applies the same logic to a regulated, high-stakes domain.

What this Shift Means for Enterprise Technology Strategy

The five forces above are not predictions. They are conditions that already exist and are already changing how enterprises make software decisions.

The $50K threshold conversation, the moment when someone at the table asks whether to build instead of buy is happening more often and resolving differently than it did three years ago. It is showing up in healthcare operations, financial services workflows, supply chain logic, and customer-facing decisioning. The contexts where custom software has shifted from a premium option to the baseline competitive requirement are multiplying.

The risk of moving too slowly is not abstract:

• Teams that built custom agent workflows before the per-seat pricing shift are not paying per-resolution fees today
• Teams that built compliance-ready AI infrastructure before August 2026 are not scrambling to certify a vendor's posture
• The gap between organisations that made the decision early and those making it now is widening every quarter

The build case, in this context, is not ideological. The numbers that made SaaS the default for fifteen years have moved. Whether they have moved enough for a specific workflow, team, and time horizon is a calculation worth running and most enterprises are not running it yet with current inputs.

How Ideas2IT Helps Enterprises Operationalize the Build Shift

The hardest part of the build-versus-buy decision is the execution.

Most teams have looked at a SaaS renewal, concluded that building was worth evaluating, and then found that the gap between that conclusion and a working production system is wider than the analysis suggested. Pricing, delivery design, team readiness, governance, and tooling all need to be figured out before a build actually compounds rather than accumulates debt. Getting any one of those wrong is the difference between a custom build that becomes a competitive asset and one that becomes a different kind of vendor dependency.

Ideas2IT's Forward Deployed Engineering model puts engineers inside the client's environment from day one working within the existing stack, standups, and OKRs rather than running a separate consulting track that produces a handover document. The team that evaluates the decision is the team that executes it, inside the organisation that has to own what gets built.

For enterprises moving to build: Anticlock is the platform that operationalises AI-driven development at team scale. The problem it addresses shows up consistently when enterprises shift from buying to building: individual developers adopt AI coding tools their own way, codebases become inconsistent at high velocity, and the internal build that was supposed to reduce vendor dependency creates a different kind of technical debt instead. Anticlock standardises AI-driven development across the engineering team enforcing consistent tooling, security guardrails, and deployment standards so every sprint cycle follows the same repeatable, auditable process. The output is at least 50% faster sprint velocity against a codebase that is maintainable by the next team that inherits it, not just the one that built it.

A Custom Software Scoping Session is the practical entry point: a two-week engagement that produces a clear build-versus-buy recommendation for a specific workflow, with scope, architecture, and delivery model defined. It is designed to answer the question that surfaces at every $50K-and-above licensing conversation with enough specificity to make a decision rather than defer one.

If your team is sitting on a SaaS renewal that crosses the threshold and someone has already asked the build question, that question deserves a real answer. Start the conversation with Ideas2IT

References

[1] BetterCloud. 2025 SaaS Management Index. BetterCloud, 2025.

[2] Vertice. SaaS Wastage: The Cost of Shelfware and Underutilized Software. Vertice, 2025. https://www.vertice.one/blog/saas-wastage-shelfware

[3] IBM Security. Cost of a Data Breach Report 2024. IBM, 2024. https://www.ibm.com/reports/data-breach

[4] Retool. The Build vs. Buy Shift: AI, Shadow IT, and the SaaS Replacement Era. Retool, February 2026. https://retool.com/blog/ai-build-vs-buy-report-2026

[5] Osika, Anton. Quoted in: "Lovable wants to be the last piece of software for companies." Fortune, December 18, 2025. https://fortune.com/2025/12/18/lovable-ai-vibe-coding-last-piece-of-software-ceo/

[6] "Cursor in talks to raise $2B at $50B valuation after hitting $2B ARR in three years." The Next Web, 2026. https://thenextweb.com/news/cursor-anysphere-2-billion-funding-50-billion-valuation-ai-coding

[7] Patel, Dylan (SemiAnalysis). Threads post, February 27, 2026. https://www.threads.com/@semianalysis/post/DVRUiXVlA7z

[8] Masad, Amjad. X post, February 4, 2025. https://x.com/amasad/status/1886516600653930924

[9] Nadella, Satya. BG2 Podcast, late 2024. Transcript at bg2pod.com.

[10] Levie, Aaron. X post, March 2025. https://x.com/levie/status/1898904236936814900

[11] Salesforce. Q3 FY26 Earnings Press Release. December 3, 2025. https://www.salesforce.com/news/press-releases/2025/12/03/fy26-q3-earnings/

[12] Chen, Joanne and Gupta, Jaya. "AI leads a service as software paradigm shift." Foundation Capital, April 2024. https://foundationcapital.com/ai-service-as-software/

[13] KPMG. AI Quarterly Pulse Survey. KPMG, June 2025.

[14] European Commission. EU Artificial Intelligence Act. Official Journal of the European Union, 2024. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai

[15] European Data Protection Board. Annual Report 2024. EDPB, 2025. https://www.edpb.europa.eu/our-work-tools/our-documents/annual-report/annual-report-2024_en

[16] Gartner. Emerging Technologies: Multicloud Architecture Adoption Drivers. Gartner, 2025.

[17] Forrester Research. Worldwide SaaS And Cloud Software Forecast, 2024 To 2029. Forrester, 2024.

‍