Coding agents lower the cost of creating and modifying software, but they do not eliminate the economics of software products. The boundary is shifting: narrow workflow tools, internal utilities and custom orchestration become easier to build, while systems of record, regulated platforms, deeply integrated applications and products with strong network/data advantages remain harder to replace. The strategic decision is moving from a binary build-versus-buy question to build, buy, blend, extend or replace—evaluated against total lifecycle cost, differentiation, control, data, integration and change velocity.
Research method
This report triangulates current primary/public-sector evidence, large-sample research and relevant industry/platform documentation. It prioritises observed behaviour over forecasts, distinguishes vendor claims from neutral statistics, and uses an explicit 8i judgment layer only where source evidence ends. The analysis is current through 15 September 2026.
1. Why the software boundary is moving
McKinsey’s 2026 State of AI survey reported that 32% of respondents said their organisations had declined to purchase at least one software product or feature because coding agents enabled them to build it internally. S1 This is an important directional signal: agentic coding is beginning to affect procurement behaviour, not merely developer productivity.
Gartner’s 2026 research has responded with explicit ‘build, buy or blend’ and SaaS-replaceability frameworks for agentic applications. S2S3 The emergence of these frameworks itself reflects a market transition: AI-generated code makes creation cheaper, but architecture, maintenance and accountability remain.
2. What coding agents actually change
- The marginal cost of prototypes and small internal tools falls sharply.
- The cost of adapting existing software decreases when agents can understand repositories and generate tested changes.
- Business users and smaller engineering teams can attempt applications that previously failed the business case.
- The supply of software increases, which weakens pricing power for simple feature bundles.
- Integration and workflow orchestration become more strategically important than stand-alone feature depth.
- The cost of ownership can still rise if generated systems create maintenance, security, architecture or dependency debt.
3. A replaceability test
| Dimension | More buildable / replaceable | More defensible to buy |
|---|---|---|
| Differentiation | Unique workflow or proprietary process. | Commodity requirement where vendor scale wins. |
| System role | Edge workflow, local utility, orchestration. | System of record or mission-critical transaction core. |
| Integration | Few stable interfaces. | Many deep integrations and ecosystem dependencies. |
| Regulation / assurance | Low consequence and reversible. | High assurance, certification, audit or liability burden. |
| Data advantage | Enterprise owns the needed data. | Vendor has unique data/network effects. |
| Change velocity | Fast-changing workflow benefits from internal iteration. | Stable standard process benefits from product maturity. |
| Ops burden | Small user base; tolerant of internal support. | 24/7 reliability, support and global compliance required. |
4. SaaS categories most exposed
The categories most exposed are those where value is concentrated in a relatively thin workflow layer: lightweight internal dashboards, approval flows, simple CRUD applications, niche reporting, data transformation, basic content operations and point solutions with limited proprietary data. These can increasingly be assembled from cloud primitives, APIs and agent-generated code.
The least replaceable categories are not necessarily the most feature-rich. They are the ones carrying institutional trust, records, regulated workflows, ecosystems, switching costs, identity, payments, security or data networks. Replacing the interface is easier than replacing the institutional function.
Sources and notes
- McKinsey, The state of AI in 2026: On the road to ROI (25 Aug 2026). [Source](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai?os=av.)) — Includes software-purchase substitution and enterprise scaling evidence.
- Gartner, Agentic Automation: Decide When to Build, Buy or Blend (8 Jul 2026). [Source](https://www.gartner.com/en/documents/8108997)) — Research abstract; build/buy/blend framing.
- Gartner, Evaluating SaaS Replaceability in the Era of Agentic Applications (14 Jul 2026). [Source](https://www.gartner.com/en/documents/8128129)) — Research abstract; SaaS portfolio replaceability framework.
8i TECHNOLOGIES
Build vs Buy in the Agentic Software Era
How Coding Agents Change SaaS Economics, Application Strategy and the Boundary of the Enterprise Software Market
Reference date: 15 September 2026
See what matters next.
| Executive answer Coding agents lower the cost of creating and modifying software, but they do not eliminate the economics of software products. The boundary is shifting: narrow workflow tools, internal utilities and custom orchestration become easier to build, while systems of record, regulated platforms, deeply integrated applications and products with strong network/data advantages remain harder to replace. The strategic decision is moving from a binary build-versus-buy question to build, buy, blend, extend or replace—evaluated against total lifecycle cost, differentiation, control, data, integration and change velocity. |
|---|
Research method
This report triangulates current primary/public-sector evidence, large-sample research and relevant industry/platform documentation. It prioritises observed behaviour over forecasts, distinguishes vendor claims from neutral statistics, and uses an explicit 8i judgment layer only where source evidence ends. The analysis is current through 15 September 2026.
1. Why the software boundary is moving
McKinsey’s 2026 State of AI survey reported that 32% of respondents said their organisations had declined to purchase at least one software product or feature because coding agents enabled them to build it internally. [S1] This is an important directional signal: agentic coding is beginning to affect procurement behaviour, not merely developer productivity.
Gartner’s 2026 research has responded with explicit ‘build, buy or blend’ and SaaS-replaceability frameworks for agentic applications. [S2][S3] The emergence of these frameworks itself reflects a market transition: AI-generated code makes creation cheaper, but architecture, maintenance and accountability remain.
2. What coding agents actually change
- The marginal cost of prototypes and small internal tools falls sharply.
- The cost of adapting existing software decreases when agents can understand repositories and generate tested changes.
- Business users and smaller engineering teams can attempt applications that previously failed the business case.
- The supply of software increases, which weakens pricing power for simple feature bundles.
- Integration and workflow orchestration become more strategically important than stand-alone feature depth.
- The cost of ownership can still rise if generated systems create maintenance, security, architecture or dependency debt.
3. A replaceability test
| Dimension | More buildable / replaceable | More defensible to buy |
|---|---|---|
| Differentiation | Unique workflow or proprietary process. | Commodity requirement where vendor scale wins. |
| System role | Edge workflow, local utility, orchestration. | System of record or mission-critical transaction core. |
| Integration | Few stable interfaces. | Many deep integrations and ecosystem dependencies. |
| Regulation / assurance | Low consequence and reversible. | High assurance, certification, audit or liability burden. |
| Data advantage | Enterprise owns the needed data. | Vendor has unique data/network effects. |
| Change velocity | Fast-changing workflow benefits from internal iteration. | Stable standard process benefits from product maturity. |
| Ops burden | Small user base; tolerant of internal support. | 24/7 reliability, support and global compliance required. |
4. SaaS categories most exposed
The categories most exposed are those where value is concentrated in a relatively thin workflow layer: lightweight internal dashboards, approval flows, simple CRUD applications, niche reporting, data transformation, basic content operations and point solutions with limited proprietary data. These can increasingly be assembled from cloud primitives, APIs and agent-generated code.
The least replaceable categories are not necessarily the most feature-rich. They are the ones carrying institutional trust, records, regulated workflows, ecosystems, switching costs, identity, payments, security or data networks. Replacing the interface is easier than replacing the institutional function.
5. The hidden cost of building
The most common analytical error is comparing SaaS licence price with coding time. The real comparison is total lifecycle cost: product ownership, architecture, testing, security, observability, documentation, data quality, upgrades, incident response, compliance, user support and staff continuity. Coding agents reduce some components; they do not remove them.
The larger the business consequence, the less sensible it is to treat generated code as a one-off asset. Internal applications require a product owner, service-level expectations and retirement logic just as purchased applications do.
6. How vendors should respond
- Move value from features toward trusted systems, proprietary data, workflow integration and measurable outcomes.
- Expose APIs and agent interfaces so customers can compose rather than replace.
- Adopt usage/outcome pricing where per-seat value erodes.
- Use AI to compress implementation and configuration time.
- Become the governed system of record or control plane underneath customer-built agentic experiences.
- Prove lower total cost and risk versus internal maintenance, not merely a longer feature checklist.
7. 8i decision framework
8i recommends a five-way decision: BUY when scale, trust or ecosystem economics dominate; BUILD when the workflow is differentiating and bounded; BLEND when a commercial system of record can be extended with custom agents; REPLACE when the purchased layer is thin and switching risk is low; RETIRE when the process itself should disappear rather than be reimplemented.
This should become a repeatable research dataset: application category, annual spend, replaceability score, AI-generated build effort, assurance burden, integration depth and observed substitution examples.
Source register
Primary and high-value sources used. Time-sensitive claims were checked against the latest available release as of the reference date. Commercial/vendor research is identified as such and is not treated as neutral market truth without corroboration.
[S1] McKinsey, The state of AI in 2026: On the road to ROI (25 Aug 2026). Source — Includes software-purchase substitution and enterprise scaling evidence.
[S2] Gartner, Agentic Automation: Decide When to Build, Buy or Blend (8 Jul 2026). Source — Research abstract; build/buy/blend framing.
[S3] Gartner, Evaluating SaaS Replaceability in the Era of Agentic Applications (14 Jul 2026). Source — Research abstract; SaaS portfolio replaceability framework.
[S4] Gartner, Build vs. Buy Application Decisions for AI Development or SaaS (11 Aug 2026). Source — Research abstract; seven-category evaluation model.
[S5] Stanford HAI, 2026 AI Index — Economy. Source — AI adoption, agent use and software-development productivity evidence.
[S6] U.S. Census Bureau, Microstructure of AI Diffusion (Apr 2026). Source — Nationally representative evidence on AI diffusion by firm/function/task.