AI for technology

AI for technology

Technology is the function that owns the code and the systems. The opportunity with AI is to compress the time from idea to production, to reduce the operational toil, and to increase the consistency of the operational work. The risk is to reduce ownership and quality in pursuit of speed.

How to think about this

Three operational zones carry most of the economic case:

• Software development. The coding agent that proposes code, runs tests, reviews pull requests, and refactors. The developer stays in the loop; the agent does the reading, the writing, and the routine. The merge is the developer’s.
• IT operations. The agent that triages alerts, gathers context from logs and dashboards, groups related signals, and prepares an incident brief. The on-call engineer takes the action; the agent does the reading and the writing of the routine.
• Documentation and knowledge. The agent that maintains the documentation, the runbooks, the architecture diagrams. The knowledge is always current; the human team reviews the changes.

What this looks like in production

The agent integrates with the existing toolchain. It does not introduce a new control plane; it works within the existing one. The agent’s outputs are reviewed by the human before they are accepted.

For software development, the agent proposes code that the developer reviews. The merge is the developer’s decision. The agent does not bypass code review; it accelerates it.

For IT operations, the agent triages and prepares. The on-call engineer takes the action. The agent does not take production actions without human approval.

For documentation, the agent maintains the source of truth. The human team reviews the changes. The agent does not silently change documented behaviour; it surfaces the change for human review.

When NOT to use this approach

• If the team is not prepared to review agent outputs before merge. The agent is the proposer; the human is the approver.
• If the production system actions cannot be human-approved. The agent must not be the only writer to production.
• If the documentation generated by the agent would be accepted without review. The agent is the maintainer; the human is the reviewer.

What good looks like

• Developer time shifted from context switching to writing code, with the agent handling the routine and the developer handling the judgement.
• Code review cycle time reduced, with the agent as the first-pass reviewer and the human as the final approver.
• IT operations triage compressed, with the agent handling the routine and the engineer handling the incidents that require human investigation.
• Documentation kept in sync with the systems, with the human team reviewing the changes the agent proposes.