Stop thinking of a workspace as a conversation. It's a room. Fill it with the right people for the job and direct them.

The single biggest workflow unlock in CharmIQ is this realization: a workspace isn't a one-on-one conversation between you and an AI. It's a collaborative space you can populate with whoever — or whatever — the work requires.

Multiple Charms. Different models. Different personas. All working in the same document, each seeing everything that came before it.

You're not a user. You're the room manager. You set the agenda, assign the work, review the output and direct revisions. The Charms are your team.

The core multi-Charm pattern is simple: tag different Charms at different points in the document, in sequence. Each Charm reads everything above it — including the output of every Charm that ran before.

The workspace accumulates the full context. Later Charms aren't working in isolation — they're building on everything that came before. That's the difference between sequential AI calls and genuine collaborative work.

When you're not sure about a Charm's output, don't just accept it — and don't ask the same Charm to check its own work. Ask a 

Tag in a Charm that would be qualified to critique the work. If a researcher Charm wrote a brief, bring in a domain expert Charm to review it:

\u I'm a little suspicious of the conclusions in section 2 — what do you make of them?

The reviewing Charm reads the original output as part of the document context and gives you an honest assessment. You get a quality check without being the only quality control in the room.

Not every Charm needs to be fast. The expensive, slower models — Claude Opus and its equivalents — are not worth running for iterative work. The speed cost is too high.

They're worth it for exactly one thing: the final review.

After you've completed a major piece of work, switch to your Opus-tier Charm and ask one question:

\u Is there anything I haven't considered?

That's the whole prompt. The depth of what surfaces from a single final-pass review at that model tier is often the most valuable thing you get from the entire session.

Use it sparingly. Use it deliberately. It's not a drafting tool — it's a finishing move.

Don't try to do everything in one giant document. A workspace that covers research, strategy, asset development and planning all at once is working against you in two ways: it creates cognitive overload for you and it dilutes the Charm's context signal.

Separate workspaces for separate phases keeps both sharp:

The pipeline: finish one workspace → extract the finished output → paste it clean into the next workspace as Reference Material. No noise travels forward. The Charm in the next phase sees only the finished product of the previous one, not the exploratory mess that produced it.

A meeting transcript is raw context. Drop it into a workspace and a Charm can do things with it that no summary document can replicate:

Try the "outsider" move: give a transcript to a Charm with no prior context about who you are or what you're building and ask it to respond as if hearing it for the first time. The critiques and observations that come back are often sharper than anything you'd generate yourself.

After your next significant meeting: drop the transcript in. Ask: 

 — Designing Charm personas that show up ready for the job.

 — Equip individual Charms with the right access before putting them to work together.

 — Iteration patterns, annotation, and when to start a new document.

 — Using the multi-Charm approach to generate and refine presentations.

How to run multiple Charms in sequence on a single problem — and why the best results come from directing a room, not managing a chatbot.

Workspace	Purpose
Charm Creation	Design and test new Charm personas
Research	Gather and synthesize source material
Strategy	Work through the problem with expert Charms
Asset Development	Execute — copy, slides, deliverables

Multi-Charm Workflows

The Room, Not the Chatbot

Sequential Tagging

Charm Checks Charm

The Opus Final Check

Partition Your Workspaces by Function

Using Transcripts

Patterns

Next Steps