# Anthropic Just Revealed How to Prompt Opus 5 > [!question] Lingering Questions > - The self-verification claim ("Opus 5 already checks and fixes its work as it goes") is sourced entirely to Anthropic, and the video's own demo never tests it — no paired comparison, no baseline. Worth running that comparison on my own workload before stripping verification instructions from any template. > - Which of the five rules are model-invariant (scope control) versus model-specific (self-verification behavior)? Prompting guidance tuned to one model has a short shelf life. > [!tip] Article Ideas > - The paired-prompt test is a same-afternoon experiment: run one task with and without "check your work," compare quality, tokens, latency. Nobody appears to have published this. > - "Add an accessibility clause to your prompt template" — none of the five rules produce accessible output, and the verification step described counts sections rather than checking heading order, alt text, or contrast. > [!link] Cross Reference > - [[3. Resources/Permanent Notes/Effort Is a Cost-Quality Dial That Subsumes Prompt Complexity]] — the vault's existing claim this source's H1 confirms from a second angle > - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] > - [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] > [!quote] Vocabulary > - **Effort level** — the setting controlling how much thinking/compute a model spends on a task; a cost/latency dial, not a direct quality dial ## Bib capsule Paul J Lipsky summarizes and demos Anthropic's published prompting guidance for Claude Opus 5 — five rules distilled from the vendor's own documentation, illustrated by building a single prompt incrementally for a fictional coffee brand's one-page website. ## Author's thesis Two of the five rules invert prompting advice that's been standard for years — most consequentially, that Opus 5 already verifies its own work as it goes, so instructing it to double-check is often redundant cost rather than added quality; the other rules (bound the job, bound the response, bound the artifact, gate clarifying questions with a threshold) are ordinary good hygiene that would help with any model. ## Key arguments (≤3) 1. **Effort is a cost dial, not a quality dial** — spending more compute only improves results when compute was the actual constraint; defaulting to maximum effort usually pays more for the same answer. 2. **Constraining the response is not constraining the artifact** — a three-bullet reply can still ship a sprawling website; these require separate instructions. 3. **The self-verification-is-redundant claim is the video's most consequential and least tested rule** — offered as vendor guidance with no independent comparison run. ## Claim atoms ### [H1] Effort level is a cost dial, not a quality dial > [!quote] ID:rw_pending > The effort setting mainly controls how much thinking Claude does, which affects the time and tokens it spends on the task. While that can produce better answers, sometimes, a lot of the time, it's just overkill. Where/why: [00:02:23], the effort-level section. **Already captured** — see [[3. Resources/Permanent Notes/Effort Is a Cost-Quality Dial That Subsumes Prompt Complexity]]. ### [H2] Gate clarifying questions with a stated threshold > [!quote] ID:rw_pending > Only ask me a question if the answer would change the whole website. Where/why: [00:04:48], rule two. The most portable single line in the video — converts "ask if unclear" from an open invitation into a decision criterion. ### [H3] Bounding the response does not bound the artifact > [!quote] ID:rw_pending > if we left the prompt as is, we might get a short response from Claude, but the website could be too large Where/why: [00:06:25]. Explains a failure people experience without diagnosing — two separate controls requiring two separate instructions. ### [H4] Opus 5 self-verifies, so re-check instructions are redundant > [!quote] ID:rw_pending > Anthropic says that Opus 5 already checks and fixes its work as it goes along. If we force it to do the whole check again, it may take more time and use more tokens without actually making the end result any better. Where/why: [00:08:00], the rule that reverses standard practice. The most consequential and least supported claim in the video — sourced entirely to the vendor, untested in the demo. ## Permanent candidates - **[P1] More computation is not the same as better output** — already exists in the vault, see [[3. Resources/Permanent Notes/Effort Is a Cost-Quality Dial That Subsumes Prompt Complexity]] (not duplicated here) - **[P2] Gate clarifying questions with a stated threshold** — promoted, see [[3. Resources/Permanent Notes/A Stated Threshold Gates Clarifying Questions Better Than an Open Invitation]] - **[P3] Self-verification claims need a paired comparison before they're trusted** — promoted, see [[3. Resources/Permanent Notes/A Vendors Capability Claim Needs a Paired Comparison Not a Demo]] ## Processing checklist - [x] Port claim atoms and permanent candidates from the existing deep-read - [x] Confirm H1 already exists as a Permanent note — connected rather than duplicated - [x] Connect into [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] and [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] - [ ] Run the paired-prompt test on the self-verification rule against a real workload ## Links - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — a prompting-practice source relevant to how professionals adapt working habits to model changes - [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] — the effort-as-cost-dial claim belongs directly in this MOC's model/effort policy tracking - [[3. Resources/Readwise/Articles/Anthropic Just Revealed How to Prompt Opus 5]] — source document - [[3. Resources/MOCs/AI Agent Safety MOC]] — applies-in: the MOC lists this note under Persistent instruction & steering.