Which software categories face declining demand versus increasing demand as…

Which software categories face declining demand versus increasing demand as Artificial Intelligence (AI) coding agents make custom software generation cheap?

2026-04-28 · agentic-ai tools-infrastructure cost-performance · medium · source → · wiki →

key claims

Commercial software categories whose value is mostly packaged business logic or user-interface convenience face the clearest near-term demand decline, because AI-assisted internal building sharply reduces the historical buy premium for those functionsRetool (2026)Simonwillison (n.d.)Ronald (1937)
Current survey evidence shows that the replacement pressure is already concentrated in workflow automation, internal admin tools, business-intelligence tools, customer-relationship management adjacencies, project management, and customer-support tooling rather than evenly across all Software-as-a-Service categoriesRetool (2026)
Concrete enterprise cases such as Klarna's internal rebuild of customer-relationship workflows suggest that firms with strong proprietary context can justify rebuilding selected applications that would previously have been bought, even if the pattern is not universalTechCrunch (n.d.)Deloitte New Zealand (n.d.)
Platform engineering and Internal Developer Platform demand rises when software becomes cheaper to produce, because the scarce asset shifts from raw coding capacity to a safe, low-friction path for many humans and agents to ship software repeatedlyESG (n.d.)Frontiers in Computer Science (n.d.)Stripe (n.d.)
Cloud hosting, Continuous Integration and Continuous Delivery pipelines, and observability gain demand because higher software volume produces more release events, more environments, more telemetry, and more operational failure modes even when generating the code itself gets cheaperDeloitte New Zealand (n.d.)Splunk (n.d.)Stripe (n.d.)
Identity-related software is likely to become more valuable, because every new application, integration, external participant, and non-human actor expands the access surface that must be authenticated, authorised, and auditedNIST (n.d.)Helpnetsecurity (n.d.)
Agent-facing interface layers, especially machine-readable Application Programming Interfaces and related tool-access surfaces, become a growth category because agents behave as software consumers that need explicit schemas, durable authentication, and predictable error handling at machine speedApify (n.d.)Stripe (n.d.)
The most durable commercial applications remain those with moats that AI-assisted building does not erase, including proprietary data, network effects, deep compliance packaging, or hard-to-replicate multi-tenant operating infrastructure, and incumbents can preserve demand by moving up into orchestration or broader suites rather than selling isolated logicRonald (1937)Reuters via Yahoo Finance (n.d.)TechCrunch (n.d.)In-parallel (n.d.)

Research Question

As Artificial Intelligence (AI) coding agents, such as Anthropic Claude Code, OpenAI Codex, and GitHub Copilot Workspace, make custom software generation materially cheaper, which categories of commercial software face declining demand because the build-it-yourself path becomes viable, which categories face increasing demand because cheaper software production amplifies the need for them, and what structural properties distinguish each group?

Findings

Executive Summary

Custom-software demand is most likely to substitute away from narrow application-layer SaaS categories, not from coordination-bearing infrastructure categories. The most exposed products are those whose buy premium came mainly from implementation effort, such as workflow automation, internal admin tools, lightweight business-intelligence tools, and some customer-relationship or project-management workflows. The categories most likely to gain demand are those that absorb scale, governance, deployment, identity, and operational complexity costs that AI coding does not remove, including platform engineering, cloud hosting, CI/CD, observability, identity, and agent-facing interface layers. This does not mean incumbent vendors or total application spend disappear, because vendors that move up the stack into orchestration, bundled suites, proprietary data, or broader workflow packaging can preserve or expand demand even while narrow logic-only tools commoditise.

Key Findings

Commercial software categories whose value is mostly packaged business logic or user-interface convenience face the clearest near-term demand decline, because AI-assisted internal building sharply reduces the historical buy premium for those functions. [inference] (medium confidence; source: Retool: The Build vs Buy Shift report 2026 simonwillison.net Ronald Coase, The Nature of the Firm, Cornell-hosted PDF of the 1937 article
Current survey evidence shows that the replacement pressure is already concentrated in workflow automation, internal admin tools, business-intelligence tools, customer-relationship management adjacencies, project management, and customer-support tooling rather than evenly across all Software-as-a-Service categories. [fact] (medium confidence; source: Retool: The Build vs Buy Shift report 2026
Concrete enterprise cases such as Klarna's internal rebuild of customer-relationship workflows suggest that firms with strong proprietary context can justify rebuilding selected applications that would previously have been bought, even if the pattern is not universal. [inference] (medium confidence; source: TechCrunch: Klarna CEO doubts that other companies will replace Salesforce with Artificial Intelligence Deloitte New Zealand: AI-assisted software engineering
Platform engineering and Internal Developer Platform demand rises when software becomes cheaper to produce, because the scarce asset shifts from raw coding capacity to a safe, low-friction path for many humans and agents to ship software repeatedly. [inference] (high confidence; source: Google Cloud and Enterprise Strategy Group (ESG): new platform engineering research report Frontiers in Computer Science: platform engineering and internal developer portals multivocal literature review Stripe: Minions, Stripe's one-shot end-to-end coding agents
Cloud hosting, Continuous Integration and Continuous Delivery pipelines, and observability gain demand because higher software volume produces more release events, more environments, more telemetry, and more operational failure modes even when generating the code itself gets cheaper. [inference] (medium confidence; source: Deloitte New Zealand: AI-assisted software engineering Splunk: the complete guide to Continuous Integration and Continuous Delivery (CI/CD) pipeline monitoring Stripe: Minions, Stripe's one-shot end-to-end coding agents
Identity-related software is likely to become more valuable, because every new application, integration, external participant, and non-human actor expands the access surface that must be authenticated, authorised, and audited. [inference] (medium confidence; source: National Institute of Standards and Technology (NIST): Identity and Access Management resource center Help Net Security summarising ConductorOne's survey on identity risks and complexity
Agent-facing interface layers, especially machine-readable Application Programming Interfaces and related tool-access surfaces, become a growth category because agents behave as software consumers that need explicit schemas, durable authentication, and predictable error handling at machine speed. [inference] (medium confidence; source: Apify: prepare your Application Programming Interface (API) for the agentic economy Stripe: Minions, Stripe's one-shot end-to-end coding agents
The most durable commercial applications remain those with moats that AI-assisted building does not erase, including proprietary data, network effects, deep compliance packaging, or hard-to-replicate multi-tenant operating infrastructure, and incumbents can preserve demand by moving up into orchestration or broader suites rather than selling isolated logic. [inference] (medium confidence; source: Ronald Coase, The Nature of the Firm, Cornell-hosted PDF of the 1937 article Reuters via Yahoo Finance: selloff wipes out nearly $1 trillion from software and services stocks as investors debate AI's existential threat TechCrunch: Klarna CEO doubts that other companies will replace Salesforce with Artificial Intelligence summary of the 2025 Artificial Intelligence deck

Assumptions

Assumption: Near-term demand shifts will show first in contract-renewal pressure, new-tool avoidance, and selective internal rebuilds rather than in a uniform collapse of incumbent software revenue. Justification: Direct evidence exists for category-level pressure and expanding internal builds, but not for a complete replacement wave.
Assumption: Control-plane categories that rise in enterprise settings will also capture value in the broader commercial market because software proliferation creates similar coordination problems outside any one firm. Justification: The available direct evidence is enterprise-heavy, so the broader-market extrapolation remains inferential.

Analysis

The evidence was weighted most heavily when it showed current category behavior rather than abstract future possibility. Retool is the strongest direct source for declining-demand categories because it identifies which SaaS classes customers are already replacing, while Google Cloud and the Frontiers review are the strongest direct sources for increasing-demand categories because they document current platform-engineering expansion rather than only arguing for it. The transaction-cost frame resolves the apparent contradiction between "software gets cheaper" and "infrastructure becomes more valuable": AI removes some production cost, but it does not remove the coordination cost of running many artifacts safely, so value migrates toward categories that absorb that coordination burden. Repository companions strengthen that reading because previous constraint-removal episodes in this corpus repeatedly shifted value from local execution toward shared platforms and control planes.

Risks, Gaps, and Uncertainties

The strongest direct replacement evidence comes from one vendor survey, so category-level magnitudes should be treated as directional rather than as market-share estimates.
Public-equity volatility captures investor fear faster than product-market reality, so valuation moves overstate the certainty of near-term disruption.
Large-firm rebuild examples may not generalise cleanly to smaller firms that lack data, engineering depth, or integration discipline.
Agent-facing interface demand is likely real but still early, so the size and timing of that category expansion are more uncertain than the expansion in platform engineering or identity.

Open Questions

Which public software vendors are already preserving demand by moving up the stack from application logic into orchestration, governance, or proprietary-data layers?
How much of the new value pool will accrue to traditional identity vendors versus new machine-identity and agent-access vendors?
At what point does software proliferation create enough maintenance burden to slow the rebound effect and favour consolidation back into shared platforms?