Make-or-buy: the economics are changing

Make becomes faster and less expensive#

Coding agents change the economics of custom software. A small team can reach a suitable solution faster and at lower cost than it could two years ago. Agentic make is not automatically better than buy, but it is economically realistic more often.

Software suppliers use the same models and tools. Their advantage lies not only in development but in scale: security, regulatory work, maintenance, and operations are spread across many customers. When standard software can be used without major customisation, buy therefore remains difficult to beat.

Capital markets showed how seriously investors take the shift in early February 2026. Anthropic had just released agentic tools that handle complete steps in legal, sales, and analytical workflows. Reuters identified that release as an important trigger for the sell-off that followed; affected companies included Salesforce, ServiceNow, and data and IT-services providers. Share prices do not prove new make-or-buy economics. They show that the market is already pricing them in.[1][2][3]

Coding speed is not yet shorter time to market#

Coding agents shorten implementation. Time to market also includes business clarification, integration, security, testing, and release. If those steps remain unchanged, even a solution developed faster will not reach users sooner.

Studies find the largest time savings in clearly bounded coding tasks. The effect is smaller in enterprise work because existing systems, coordination, and quality assurance enter the picture. Make-or-buy should therefore not compare coding output. It should compare the time and cost from idea to production change.

Buy often appears more concrete during selection because a supplier can demonstrate a finished product. At that point, make usually exists only as a cost estimate. Coding agents reduce that advantage because an agentic-make prototype can address the same use case with the same data and requirements early. The enterprise can then compare two solutions instead of a product with a presentation.[4]

Time to market follows the slowest feedback loop#

Once implementation accelerates, unresolved requirements account for more of the elapsed time. The team is no longer waiting for code but for decisions about exceptions, priorities, and intended behaviour. Coding agents do not make requirements less important; open questions simply slow the work sooner.

AI can bring existing process knowledge together, flag contradictions, and draft a specification. Agent skills structure open questions, maintain domain language, and turn decisions into small, testable work packages. Without the business objective, stakeholder context, and relevant edge cases, the result remains generic. The business must still decide how the process should work.

The open-source Agent Skills Toolbelt bundles skills for requirements clarification, debugging, and review.[5][6][7]

Validation must scale with output#

After business clarification comes the next feedback loop: review. In a study covering 802 developers, the number of submitted code changes per developer doubled. Reviewer load roughly doubled with it. The case does not transfer to every enterprise, but it exposes the mechanism: when code is produced faster than review scales, the time saved in implementation is lost downstream.

Automated tests and reviews are therefore necessary but insufficient. If code and tests come from the same ambiguous specification, both can confirm the same error. Business-critical logic needs an independent reference, such as approved examples, regulatory rules, or production data.

Not every change needs the same process. A new billing rule requires more validation than a reversible change to an internal interface. More code is not the objective. What matters is releases that are correct for the business and work reliably in operation.[9]

Custom software means product ownership#

Even with coding agents, make is more than a one-off project. The enterprise takes responsibility for product decisions, maintenance, and operations. In return, it can change workflows, interfaces, and the use of data without waiting for a supplier's roadmap.

The enterprise must control its code and data, deploy changes itself, and operate the product. It also needs a permanently accountable product team. If that freedom creates no measurable value, the scale advantage of buy still wins.[10]

Which initiatives should be reassessed today#

Agentic make does not make buy inherently worse. It moves the point at which custom development becomes more economical. Enterprises should therefore reassess initiatives between the previous and the new make-or-buy crossover.