There is a meaningful shift happening in how organisations use AI for operations, and it is easy to miss if you only look at the tools. The shift is not technological, it is relational: AI is moving from assistant to delegate. An assistant helps you do your work.
A delegate does the work, and you review it. Most companies are still mentally in the assistant era while the technology has already crossed into delegation. The two domains where I see this most clearly are software development and content management — and together they offer a playbook for automating almost any operational function.
Start with coding, because it is furthest along. The first wave of AI in development was autocomplete — useful, incremental, unthreatening. The current wave is different. Teams now hand entire tasks to AI agents: build this integration, refactor this module, write the tests for that service, investigate why this job fails at night. The agent works through the task across multiple steps and presents a result.
The human role compresses into two activities: specifying the task well, and reviewing the output critically.
This compression changes what skill means. The developers who extract the most value are not the fastest typists; they are the clearest thinkers. A vague instruction to an AI agent produces confident garbage at remarkable speed. A precise specification — what to build, what constraints apply, what good looks like — produces work that needs only refinement. Specification and review are becoming the craft. Production is becoming the commodity.
Content management is following the same curve, slightly behind. Consider what a content operation actually consists of: drafting, editing, tagging, categorising, translating, reformatting one asset into a dozen channel-specific variants, refreshing outdated pages, maintaining consistency of tone across hundreds of pieces. Almost every step is now automatable as a pipeline. A single source piece can be transformed into a newsletter section, several social variants, a translated version, and a summary — with metadata applied and publication scheduled — before a human looks at any of it. The human looks at the end, as an editor-in-chief, not at every step as a producer.
From watching both domains, I have distilled a few principles that I believe generalise to any operational automation.
Automate the production, never the accountability. Someone with a name must own everything that ships — every deployment, every published page. The moment output goes live because no human felt responsible for stopping it, you have built a liability machine. Review gates are not inefficiency; they are the design.
Automate workflows, not tasks. Replacing one manual step with AI while the surrounding process stays manual yields almost nothing — the bottleneck just moves. The gains come from redesigning the chain end to end around the assumption that machines do the repeatable middle.
Start where failure is cheap and private. Internal automation — code generation, test creation, content drafting, documentation — fails safely. You catch the errors in review, you learn the failure patterns, you build organisational judgment about what AI can and cannot be trusted with. Only then extend automation toward customer-facing surfaces, where mistakes are public and expensive. Companies that do this in reverse order learn the same lessons at much higher tuition.
Measure outcomes, not activity. The point of automating content operations is not producing ten times more content; the internet does not need it. The point is freeing human attention for the work machines cannot do — judgment, strategy, the few pieces that genuinely require a human voice. If your automation metrics celebrate volume, you have automated yourself into noise.
Finally, expect the workforce question and answer it honestly. Delegation-level AI does reduce the hours needed for production work. Pretending otherwise insults people's intelligence and breeds resistance. The honest framing is that roles shift toward specification, review, and exception handling — and the organisation must invest in moving people there, not just hope they adapt.
The horizon question I am watching: how far up the chain does delegation climb? Today AI drafts the code and the content while humans hold the judgment. The moment AI begins reliably reviewing its own kind of work, the human layer compresses again. I do not think we are there. But I no longer think it is far — and the operations leaders who thrive will be the ones who redesigned their teams before that moment, not after it.
The above reflects my personal views only and is intended for informational and discussion purposes. It does not represent the position of any employer or organisation.