Should you still buy business software, or have an AI agent build it for you? According to McKinsey's global "The State of AI in 2026" survey, published in early September 2026, 32% of organizations have already skipped buying at least one software product because they could build it themselves with agentic coding tools. For an SME comparing SaaS subscriptions against in-house alternatives every month, this number deserves a closer look.
In brief
- 32% of organizations skipped buying at least one software product in 2026 thanks to agentic coding tools, according to McKinsey (survey fielded May 4 to June 8, 2026, 1,719 respondents in 97 countries).
- The technology sector leads (41%), followed by healthcare (39%) and professional services (38%).
- Among the highest-performing AI companies ("high performers"), nearly 50% skip software purchases, versus 31% for everyone else.
- The share of large enterprises scaling AI agents in at least one function rose from 27% to 40% in one year.
- The other side of the coin: 1 organization in 5 already says it is limiting AI use because of operating costs.
The "build vs buy" shift documented by McKinsey
Build vs buy is a classic business trade-off: develop a tool in-house, or subscribe to an existing piece of software. What is changing in 2026, according to McKinsey, is the cost of the "build" option. Agentic coding tools, meaning AI agents able to write, test, and fix code largely on their own from natural-language instructions, are making in-house development accessible to teams that would never have considered coding their own tool two years ago.
The survey, conducted online from May 4 to June 8, 2026, among 1,719 respondents in 97 countries, measures this shift precisely: 32% of organizations skipped buying at least one software product because they could build it in-house with these tools.
A clear gap between the most advanced companies and the rest
The shift is not uniform. McKinsey distinguishes "high performers", the 6% of organizations that attribute at least 5% of their EBIT to AI, from the rest of respondents. Among these most advanced companies, nearly half already skip software purchases thanks to agentic AI, compared with 31% for everyone else. In other words: the more an organization masters AI, the more it shifts toward "building".
This gap also shows up in large-scale agent deployment. Among companies with more than $1 billion in annual revenue, the share running AI agents "at scale" in at least one function rose from 27% to 40% in a single year.
An SME obviously does not have a multinational's budget. But the logic transfers: the more a team builds skill with agentic tools, the more it can consider building a tailored tool instead of paying for a generic subscription that is a poor fit for its needs.
Buy or build: how an SME should decide
The question is not "buy or build" as a binary choice, but when each option becomes the right one. Standard software often remains the right choice for generic needs already well covered by the market (accounting, payroll, a standard CRM). Building becomes attractive for a specific need poorly served by existing offerings, or for a small internal tool with low security stakes.
Buy software (SaaS)
Fast setup, vendor support included, automatic updates. Recurring cost per user, features sometimes oversized or a poor fit, and dependence on the vendor's roadmap.
Build with an AI agent
A tool tailored exactly to the need, no recurring subscription once built. Requires a minimum of in-house skill to scope, test, and maintain the tool, plus vigilance on the security of the data it handles.
The table below summarizes the criteria to weigh before deciding.
| Criterion | Buy (SaaS) | Build (AI agent) |
|---|---|---|
| Time to launch | Fast, often a few days | Varies, depends on the tool's complexity |
| Cost | Recurring subscription per user | One-off development cost, then maintenance |
| Customization | Limited to the vendor's options | Tailored precisely to the business need |
| Maintenance | Handled by the vendor | Falls on the SME (updates, bugs, security) |
| Risk if it fails | Low, can switch vendors | Higher if the tool becomes critical with no backup skill |
The limits to know before jumping in
The enthusiasm for "building" has a downside McKinsey does not hide. One-fifth of surveyed organizations already say they are limiting their AI use because of operating costs, a share that stays consistent regardless of company size. Building a tool with an AI agent is therefore not free: it takes time to scope, iterate, and test, and above all requires ongoing maintenance, which is often underestimated.
Key takeaway
An in-house tool built with an AI agent has no support desk to call when it breaks. Before dropping a market software product, an SME should make sure it has, in-house or through a trusted provider, the skill to maintain the tool over time, particularly on data security.
Another point of caution: a tool built in-house to handle customer data or sensitive information must meet the same security and compliance requirements (GDPR in particular) as purchased software. "Building" does not exempt anyone from legal obligations.
What this concretely means for your SME
Three practical takeaways follow from this survey for an SME. First, map existing SaaS subscriptions and identify the ones covering a simple, repetitive, well-defined need: these are the best candidates for replacement by a lightweight in-house tool. Second, do not start without a minimum of scoping skill, in-house or via a trusted provider, to avoid a tool "hacked together" with an AI agent becoming a technical weak point. Third, address operating cost from the start: an AI agent billed by usage can quickly cost more than expected if usage spikes, a point McKinsey already identifies as a real constraint for one organization in five.
With measured optimism, this shift documented by McKinsey confirms an underlying trend: agentic tools are making software development accessible to teams that would never have considered this option before. It is now up to each SME leader to assess, function by function, whether buying remains the safer choice or whether building has finally become realistic.
FAQ
What is "build vs buy" applied to agentic AI?
It is the trade-off between buying existing software (SaaS) and building it in-house with agentic coding tools, AI agents able to write and test code largely on their own from natural-language instructions.
Does an SME really benefit from building its own tool instead of buying software?
It depends on the need. For a generic need well covered by the market (accounting, standard CRM), buying often remains safer. Building becomes worthwhile for a specific need poorly served by existing offerings, provided the SME has the skill to maintain the tool over time.
What is the main risk of building a tool with an AI agent?
The main risk is ongoing maintenance: unlike purchased software, an in-house tool has no vendor support. The operating cost of an AI agent billed by usage can also climb faster than expected, a constraint already cited by one organization in five according to McKinsey.
Where does the 32% figure for companies skipping software purchases come from?
It comes from McKinsey's global "The State of AI in 2026" survey, conducted online from May 4 to June 8, 2026, among 1,719 respondents in 97 countries, across all sectors and company sizes.
To go further on AI tools and practical adoption in business, see our guide How to choose your AI model in 2026, our article on AI agents in the enterprise, and all our AI resources for business.
