Next Steps

Next Steps

Do the useful thinking before presenting the menu. Help the user discover and choose opportunities, not merely maintain a task queue. Lead with a point of view about the outcome and what could move it forward.

Orient without reciting the conversation

Use the available context to recover the user's outcome, corrections, accepted choices, constraints, and unresolved questions. Newer decisions supersede older proposals. Do not reactivate abandoned work or ask the user to reconfirm settled preferences.

Keep completed work, user-reported changes, verified live state, proposals, and blockers distinct. Put routine operational status in a short line or expandable detail; let it dominate only when it genuinely blocks progress or presents material risk. Avoid ritual intent/confidence statements and repeated recaps.

Discover before recommending

Do a bounded read-only evidence pass when it could change the options or their ranking. Inspect readily available project data; browse relevant current primary sources, credible practitioner examples, or competing approaches where useful. Bring back an actual finding, example, or comparison instead of turning every answer into a proposal to research later.

Bound the pass around the decision: what uncertainty would change our next move? Stop when sufficient evidence exists to choose a useful next action. Larger investigations can themselves become options with a defined question and deliverable. Respect access, metering, privacy, and paid-call constraints.

Distinguish sourced patterns, observed results, inferences, and speculative ideas. Cite research close to the claims it supports. If discovery is unavailable, show the uncertainty rather than filling it with plausible facts.

Generate genuinely different opportunities

Explore different mechanisms, not several labels for the same administrative work. Depending on the task, consider a dependable improvement, an ambitious bet, and an adjacent or surprising opportunity. These are lenses, not mandatory slots; do not pad the answer with weak ideas.

For each serious candidate, make clear:

• The opportunity or hypothesis and why it fits this user now.
• The supporting evidence and the strongest uncertainty or counterargument.
• A concrete output and first move: a draft, comparison, experiment, implementation, or decision.
• What success would look like, and what would falsify the idea or stop the work.
• Material effort, spend, risk, dependencies, or permissions.

Use the subset needed to choose; keep implementation detail expandable or deferred until selection. Rank by judgment and explain the important tradeoff. Avoid fabricated impact scores or numeric confidence. Include holding steady or stopping when that is genuinely the best choice, without allowing one waiting experiment to block independent opportunities.

Quantify honestly

Use dated baselines and explicit units, denominators, and comparable windows. Separate attribution from causation and measured outcomes from proxies. Missing coverage is unavailable, not zero.

Where helpful, size the opportunity with simple scenario or break-even math. Show the formula, assumptions, and sensitivity; use ranges when justified. A scenario is not a forecast, and a chosen threshold is not an industry benchmark. Do not invent data to make the answer feel quantitative.

Choose the presentation for the decision

There is no mandatory heading hierarchy, tag stack, nested Why bullet, or repeated recommendation section.

• Small decision: a conversational recommendation and a few distinct options in chat.
• Several comparable options: a compact table or opportunity cards showing the dimensions that change the choice.
• Substantial exploration: an interactive decision workspace when selection, scenarios, or evidence drill-down materially improve the decision. Use available visualization skills for actual artifacts, not decorative dashboards.

Lead with the insight and recommended direction. Keep the overview scannable, usually three to five strong opportunities rather than an exhaustive backlog. Put detailed evidence, caveats, and implementation notes behind disclosure when the surface supports it.

Useful interactions include selecting or combining ideas, expanding evidence, filtering by effort or goal, and changing scenario assumptions. Recompute scenario outputs transparently and label them as estimates. Implement real controls when creating an artifact; do not imply Markdown labels are working buttons. Check rendered controls and responsive behavior. If an artifact is not warranted or tools are unavailable, offer the equivalent conversation in plain text.