Skip to main content

When to Start a New Thread vs. Continue an Existing One

How thread management affects AI Assistant response quality, and when to start fresh versus keep iterating in the same conversation.

Written by Engineering Swantide

Thread management in Swantide's AI Assistant directly affects the quality of responses you get. Each thread carries its own conversation history as context, so keeping threads focused leads to sharper, faster answers.

"Always start a new chat for a new topic. When asking about the same topic, keep prompting follow-up questions to dig deeper."

Start a new thread when…

  • You're switching topics. Asking about Flows, then about Apex? Start fresh.

  • You're working on a different component. Each flow, class, or object deserves its own thread.

  • You're generating a new deliverable: new test script, new solution design, new documentation.

  • Responses are getting slow. Long threads accumulate conversation history and slow down processing.

  • You're getting irrelevant responses. The AI may be carrying stale context from earlier in the thread.

Continue the existing thread when…

  • You're iterating on the same topic: refining a solution design, adding test cases to a script, or drilling deeper into the same component.

  • You're asking follow-up questions: "What about edge cases?" or "Can you also check the related Apex triggers?" Multiple follow-ups on the same topic usually beat one big front-loaded prompt.

  • You want the AI to remember previous context. If you've already explained your requirements, continuing the thread avoids repeating yourself.

Why this matters

The AI Assistant uses your current thread's conversation history as part of its context. A long thread with mixed topics can cause:

  • Context pollution: the AI references information from an earlier, unrelated question.

  • Slower responses: more conversation history means longer processing time.

  • Less accurate answers: the AI may get confused by conflicting context across topics.

Best practice: one thread per task

Think of each thread as a focused work session. Examples:

  • Thread 1: "Analyze the Lead_Assignment flow." then follow up with edge cases, risk-rating, and refactor implications in the same thread.

  • Thread 2: "Generate UAT test scripts for the Case_Escalation feature." then iterate on negative cases.

  • Thread 3: "Why is the Require_Phone_on_Lead validation rule firing for inside-sales reps?" then ask follow-ups about the related profiles and bypass conditions.

  • Thread 4: "Design a solution for the new Renewal reporting requirement." then iterate on the design across multiple prompts.

If you can't describe a thread's purpose in one sentence, it's probably too broad, so split it.

A note on permission and security audits

The Assistant's permission and security audits have improved and surface strong, well-reasoned findings. Run the audit in its own focused thread, and for completeness-critical work confirm the full list in the Data Dictionary, which is the source of truth for a definitive permission inventory. Keeping the audit in its own thread also makes it easy to re-run cleanly after a change.

Sample AI Assistant prompts

These work best as the opening prompt in a fresh thread, so the AI starts with a clean context:

  • "Analyze the flow Lead_Assignment_Process. Summarize the trigger, decisions, and actions, then identify any risks for an upcoming refactor." single named component; rich follow-up potential ("What happens for international leads?", "Can you generate a UAT script?").

  • "Generate a UAT test script for the validation rule Require_Phone_on_Lead, covering happy path and three negative cases." clear deliverable; follow up with "Add a case for inactive users" or "Now reformat as Gherkin."

  • "Explain why the Sales_Manager profile has Modify All Data, and tell me whether that's appropriate for this org." single named profile, scoped permission question. Follow up with "What would I lose if I removed it?" or "Which permission sets give the same access?" (for completeness-critical work, confirm the full list in the Data Dictionary).

Tips

  • When in doubt, start a new thread. It's always safer than continuing a stale one.

  • Don't worry about losing org context. The AI always has your full org metadata available: it queries the org's metadata with tools on demand, so starting a new thread only resets the conversation history, not the underlying metadata.

  • Name threads mentally. A one-sentence description of the thread's purpose is a good gut check that you haven't drifted off-topic.

  • Within a topic, keep digging. A short conversation of well-aimed follow-up questions usually beats one giant first prompt.

Related Articles
Did this answer your question?