The Input Layer
- Source
- x.com
- Author
- Zack Shapiro
- Date
Context
A useful AI briefing conveys the decisions that a polished output cannot reveal. Lawyer Zack Shapiro describes a prompting practice built on assignments that specify a client's priorities, acceptable compromises, relationship history and desired tone.
Shapiro distinguishes those case-specific instructions from reusable procedures that describe the order of analysis and the decisions to consider. That separation gives practitioners a way to reuse their working method while supplying fresh facts for each assignment. The transferable point is specificity about objectives and judgment, rather than a target prompt length: a long briefing earns its space when it conveys information the task actually needs.
- A Microsoft survey of 31,000 workers ranked 'prompt engineer' second-to-last among roles companies plan to add, and Fast Company found 68% of firms treat it as basic training. The job title died; the author argues the didn't.
- The mechanism behind bad output: a vague instruction makes the model regress to the mean of its training data, which is the average of the internet. That average has a recognizable texture, and people call it AI slop.
- Role-priming is the part that genuinely expired. Reasoning models no longer need 'you are a world-class expert in securities law' because they already reason well about securities law.
- What no model can infer is the situation you own: your client's risk tolerance, the counterparty's negotiating posture, which three of fourteen contract issues actually matter. Supplying that replaced the old tricks.
- agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
“My best prompts are around two thousand words long.”
“A model that can think doesn’t automatically know what to think about.”
“Large language models are functionally very literal genies.”
“In classic genie stories, the danger is never that the genie is weak, it’s that the wisher is lazy.”
Checking sign-in…
Loading comments…