Managers and AI prompting — a social thread
Social posts argue that management sharpens communication skills like stating outcomes over solutions, and that those skills improve how teams prompt and govern AI tools. Other posts show early experiments with AI agents reviewing AI-generated code, suggesting teams are testing guardrails where agents check other agents. (x.com) (x.com).
Managers are being recast on social media as prompt engineers for the workplace: people who define outcomes, constraints, and review loops for artificial intelligence systems. (openai.com) OpenAI’s prompt engineering guide says stronger results come from clear, specific instructions, while its Codex best-practices page tells teams to spell out constraints and define what “done” means before an agent starts work. (openai.com 1) (openai.com 2) Anthropic makes a similar case in its developer materials, describing prompt engineering as writing and organizing instructions for better outcomes and “context engineering” as the next step: curating the right information around the model during inference. (anthropic.com 1) (anthropic.com 2) That framing lines up with a management habit older than chatbots: telling a team the goal, the limits, and the standard for success instead of prescribing every step. OpenAI’s Codex guidance uses that exact structure in software work, with sections for constraints and completion criteria. (openai.com) The social posts pushing this idea are landing as companies move from one-off prompts to agent workflows that can plan, call tools, and hand work back for review. Anthropic’s engineering blog says agent systems depend on managing the information and instructions that surround a model, not just writing a single clever prompt. (anthropic.com) A second strand of the discussion is about oversight: using one artificial intelligence system to inspect code written by another, with a human still deciding what ships. GitHub’s documentation says reviewing code generated by GitHub Copilot, ChatGPT, or other agents is becoming a standard part of modern software work. (github.com) GitHub’s guide does not pitch fully automated approval. It recommends human oversight, testing, and structured review prompts, especially for large pull requests and older codebases where generated changes can hide subtle bugs. (github.com) Vendors are already building around that bottleneck. Microsoft said in a September 2025 engineering post that its internal artificial intelligence code review assistant was supporting more than 90% of pull requests across the company, affecting more than 600,000 pull requests per month. (microsoft.com) Google published a codelab for a production code review assistant that uses a multi-agent setup with deterministic tools such as tests and linters, rather than relying only on a language model’s judgment. (google.com) The thread running through both debates is narrower than the hype: teams are not just asking who can write the best prompt. They are testing who can specify the job, supply the right context, and build a review step strong enough that one agent can check another without removing the human from the loop. (openai.com) (github.com)
