The AI war between Anthropic and OpenAI have produced Codex via Claude Code, which is a coding tool so spectacular, that it is allowing teams, including ours, to increase ambition so far beyond what was previously possible that it is rapidly transforming the software industry beyond recognition.

This change and empowerment is forcing us to reimagine roles in software engineering organizations. Ops engineers are particularly well suited to using AI agents for innovation, given their systems and architectural knowledge. It’s challenging for productive senior developers to embrace AI coding, given their muscle memory, but when able, Codex and Claude Code are spectacular enablers. Our QA Engineers at Defiant Inc are writing powerful utilities and seem to have no problem getting the most out of terminal AI coding agents.

The skillset required to get the most out of Codex and Claude Code are systems knowledge, architectural knowledge (how things hang together), dev management, dev methodology, understanding test driven development, QA methodologies and being an excellent communicator.

There’s another component, which is the role of AI industrial psychologist. You need to know how to talk to these things. For example, have it write a markdown document with your product design and iterate on that. Then have it write integration tests before you write a line of code. Then have it reflect on what it learned from the integration tests and have it rework the product document. Then have it create a PLAN.md in the base of the project with a step by step implementation plan, and update AGENTS.md (or CLAUDE.md) to track progress in that doc. AI industrial psychology at work.

I’ve come up with a term in our team to refer to those particularly good at bringing these skills together and getting productive shippable product out of AI agents: Operators.

I like the word ‘Operator’. My favorite context is the way I heard it used at Suits & Spooks in DC, which is a conference that used to bring the intelligence community and tech community together. Someone from a 3 letter agency referred to special forces personnel as operators. It sounds badass, but it also I think captures the role of the operator is in the field. They are backed up by a team, and are bringing together a range of skillsets to accomplish a mission. They are also very mission and outcome focused.

I like how Operator places the emphasis on mission and outcome, because AI agents are particularly bad at completing the second 90% of a project, and an Operator is particularly good at driving the agent across the finish line.

A few of my own techniques are to ask it what is left until the project is complete. Simply populating the context window with that already helps. Then have it update PLAN.md , and use a prompt that uses words like “finish line”, “outcome focused”, “deadline” and so on. Drive it across the finish line. AI Agents will find ways to fill an infinite project schedule. Kinda like humans will.

We even have our own Operator Slack emoji now, which is of course Tank answering his headset in the original Matrix (1999) with: “Operator”.