The Build-vs-Buy Map
Code copilots show how quickly a build-or-buy decision can change. In 2024, engineering teams expected to build them around a single foundation-model call. Peer teams are now buying them: vendors absorbed model improvements faster than internal wrappers could be rebuilt, and maintaining a homegrown copilot cost more than the renewal it was meant to replace. Other categories are moving toward internal builds. The registry records both directions, so teams need to revisit the decision by category.
AI tools tracked in the Airframe registry. Each build option in this issue is assessed against those tools for redundancy, fit, and substitutability.
Deployment case studies indexed in the Airframe corpus, recording what teams shipped and how their chosen path held up through the next renewal cycle.
Software categories covered by the registry. Teams are revisiting build-or-buy decisions across these categories as the assumptions of three years ago change.
Vendor sponsorships, placement fees, or category-inclusion payments. Airframe applies the same independent editorial approach to every category in the corpus.
Why the choice changed.
For four decades, the inputs favored buying. Wave IV has changed three of them at once, leaving procurement and platform teams with a different decision to make.
Buying made sense when skilled engineering was expensive, integration was slow, and a SaaS contract supplied a roadmap the customer did not have to staff. All three inputs have changed in the past three years. With a foundation model, an agent harness, and a small team, an internal tool's first version can now ship in the time previously spent negotiating an order form. Its inference costs fall over time instead of rising at renewal.
Vendors benefit from the same model improvements. Categories that looked easy to build in 2023, sometimes as a weekend wrapper around one model call, had credible vendor cohorts by 2026. Those vendors can absorb improvements faster than internal teams can re-implement them. Building is feasible in more categories, but feasibility alone does not settle the choice. Teams need to assess which path works now; a 2024 reference point can give them the wrong answer.
Start with what the registry records about peer deployments. A team working without that evidence cannot see which path its peers shipped, which failures led them to reverse course within a year, or which vendors have already absorbed the model improvements the team hopes to capture by building.
For the first time in forty years, the cheapest path is the one you build. The second-cheapest is the one your peers already built and shared.
Four lenses for the decision.
Read the category through four lenses. Each supplies different evidence and changes at a different pace.
Category fracture
More than a hundred plausible vendors, with no consolidation, indicates that the category is still defining itself. Buying ties the customer to one vendor's evolving roadmap. Where an already fractured category keeps fracturing, a small internal build can be more reliable: it can adopt the same model gains without paying for each through vendor renewals.
Time to production
Builds that reach production within a quarter behave differently at renewal from builds that take more than half a year. Teams extend, harden, and reuse short builds as they learn the problem. Long builds tend to be reversed, sometimes within the same renewal cycle, when vendors developing in parallel ship equivalent features and absorb the model gains without the customer's internal carrying cost.
Peer reversals in both directions
The registry records teams that built in 2024 and bought again by 2026 because vendors caught up. It also records teams that bought in 2024 and built replacements by 2026 because the vendor fell behind the model. These reversals help assess whether a category currently favors building or buying. Each records a path another team chose, operated for a cycle, and changed after reviewing its own data.
What the foundation model contributes
When a foundation model does most of the work, vendor margins are thin and the build is shorter. When the model sits alongside proprietary data, learned routing, and substantial integration work, vendor margins are real and an internal build can underestimate the work the vendor carries. In most categories, the product extends beyond the model. Assessing that additional work helps settle the build-or-buy choice.
Illustrative reversals
The categories and reasons below illustrate recurring reversal patterns. They are not measured counts or ranked shares. Across the sixty-nine categories scored for this issue, the directional split is roughly thirty-eight favoring buy, twenty-two hybrid, and nine build. The split shows why one default answer no longer covers every category.
Use peer evidence before the renewal.
Teams often make build-or-buy decisions slowly, with limited evidence, then carry the consequences through a renewal cycle. The registry brings peer deployments into that decision.
Inside a Fortune 1000 organization, a build-or-buy meeting can pit a vendor pitch against an engineer's confidence, with too little time to assemble peer evidence. The same pattern hurt incumbents during the SaaS-to-cloud-native transition: the signal arrived years before individual teams assembled a map that reflected it.
Airframe cohort members can request a Build-vs-Buy Briefing on rotation, alongside stack audits, vendor studies, and renewal briefings. The briefing assesses one decision the operator faces this quarter against the registry, using qualitative scores. It includes what peers shipped, the categories and decision windows, reversals and reason codes, with peer evidence in the report and sources on the last page.
A quarterly registry review can replace the evidence-gathering work that once took a year of meetings.
We are watching hybrid categories for a change next quarter. Internal search and retrieval appears to be moving from build toward buy as vendors close the integration gap. Customer support routing appears to be moving from buy toward build as per-seat pricing fails to reflect the model gains operators see in their logs. Sales enrichment and BI copilots remain close calls. If model improvements continue at the pace of the last two cycles, both tip toward build by the next issue. If a category leader adds substantial proprietary data depth first, both stay bought. We will assess those changes against the next quarter's registry.