AI agents: what it means for a company that buys development
Gartner expects AI agents to reach a large share of enterprise software and to start working as teams. What to ask anyone who sells you development with agents.
What Gartner says
In an August 2025 press release, Gartner predicts that up to four in ten enterprise applications will include task-specific AI agents in 2026, up from fewer than one in twenty at the time. The same release separates an assistant, which depends on a person, from an agent, which can carry out a task end to end.
Gartner also warns about a common mistake: calling plain assistants agents, which it labels "agentwashing". And it projects that by 2027 a third of agentic AI implementations will combine agents with different skills to handle complex tasks.
What it means if you buy development
The release is about enterprise applications in general, not development teams in particular, so we should not read more into it than it says. The direction is clear, though: when someone offers you development with agents, separate the label from the practice. Three questions help:
- What does each agent do, and what does a person do? An assistant that suggests code is not the same as an agent that implements a whole task.
- How is the output reviewed? If several agents collaborate, you need to know who verifies the result before it reaches your repository.
- What evidence do you get every week? Approved specs, reviewed pull requests and visible risks, not just a promise of speed.
How we do it at Medrar
In Medrar Squads, each engineer works with their own agent cell (specifier, implementer, QA and security reviewer) inside your repository, and nothing is merged, deployed or deleted without a person approving it. We start with a 2-week Sprint Zero on a real problem of your team.
Sources
- Gartner, nota de prensa sobre agentes de IA por tarea en aplicaciones empresariales / Gartner press release on task-specific AI agents in enterprise applications(published 2025-08-26; accessed 2026-11-07)