Why the Agent Panel exists
Most people meet generative AI through a blank chat box. That works for small, well-scoped questions — but it forces you to do three jobs at once: describe your goal, choose the right model, and shape the output for the platform you publish to. Every step is a chance to get worse results.
The EXXY® AI Studio Agent Panel removes that overhead by introducing a higher-level concept: a Task. A Task encapsulates what you want to produce, which specialist agent should produce it, and what shape the result should arrive in (an Instagram caption, a LinkedIn carousel script, a PowerPoint deck, a Word document, a PDF report, and so on).
You pick the Task. The orchestrator handles the rest.
The four-layer model: Tasks, sub-tasks, Jobs, and MCP servers
Internally, EXXY® AI Studio organises work in four layers. You only ever interact with the first one.
- Task — the user-facing intent ("Create a LinkedIn post promoting our new case study", "Summarise this BIM model into a 5-slide review", "Draft a research note in IEEE style"). Every Task lives inside a category and is wired to a recommended specialist agent and one or more output formats.
- Sub-task — the orchestrator decomposes a Task into the discrete steps required to fulfil it (intake, research, drafting, refinement, formatting, export).
- Job — each sub-task becomes one or more concurrent Jobs that run on the right model at the right tier. Jobs are tracked end-to-end so you can see progress and re-run individual steps without restarting the whole Task.
- MCP server — when a Job needs a specialised capability (real-time web search, BIM file inspection, slide rendering, citation lookup, image editing, transcript diarisation), it calls an EXXY-built MCP (Model Context Protocol) server as an auxiliary tool. MCP servers are isolated, versioned, and reusable across agents, which is what lets the platform combine deep specialisation with consistent quality.
You do not need to wire any of this together. Choosing a Task wires it for you.
The specialist agents
The orchestrator (EXXY® AI) routes Tasks to one of several specialist agents based on the Task's category and the user's tier:
- Agent No. 3 — Free-Form for general writing, summarisation, translation, social posts, and everyday explainers. Your default for prose-heavy work.
- Agent No. 5 — BIM for AECO use cases: model walkthroughs, clash narration, specification drafting, compliance prep. Pairs with the BIM analysis MCP server.
- Agent No. 7 — Slides for presentation generation, redesign, and rewrite. Pairs with the slide rendering MCP server and the brand-token store.
- Agent No. 9 — Research / STEM for long-form research notes, technical write-ups, structured citation gathering, and equation-heavy work. Pairs with grounded web search and document-intelligence MCP servers.
Each agent ships with curated system instructions, a recommended model on each tier (Free vs Pro), and a vetted toolset. Tasks are pre-mapped to the right agent so you never have to guess.
Output formats — the same Task, reshaped per platform
Every Task exposes one or more output formats. The same underlying intent ("announce our new product feature") can be rendered as:
- Social posts — Instagram caption + first-comment hashtags, LinkedIn carousel script (10 frames), TikTok / Shorts hook + script, X (Twitter) thread.
- Documents — Word (.docx) report, PDF brief, plain-text email body, Markdown for newsletters.
- Presentations — PowerPoint (.pptx) or Reveal.js web deck, including speaker notes.
- Visuals — generated cover images sized for the chosen platform (1:1, 4:5, 9:16, 16:9).
Switching the format does not just change the file extension — it re-runs the relevant sub-tasks so the style, length, tone, and visual layout match the destination. A LinkedIn carousel does not look like a PDF report, and a TikTok hook does not look like a research abstract. The Agent Panel makes both available from the same Task.
How a Task runs end to end
A typical run looks like this:
- You pick a Task from the Agent Panel ("LinkedIn carousel from a case study").
- You attach or paste the source material (a document, a link, a meeting summary, a previous deck).
- The orchestrator selects the right specialist agent and decomposes the Task into sub-tasks.
- Each sub-task spawns Jobs. Jobs that need extra capability call MCP servers (search, BIM, slides, document intelligence).
- You see a live plan with each Job's status. You can edit individual Jobs ("rewrite frame 4 in a more conversational tone") without re-running the whole Task.
- You pick the output format. The platform reshapes the result for that destination and gives you a preview, an editable view, and a one-click export.
Free vs Pro: the same Tasks, different model tiers
The Task catalogue is the same on every plan. What differs is the model tier the orchestrator is allowed to dispatch to, and the parallel Job concurrency. Pro plans run on faster reasoning and image models, allow longer context per Task, and unlock additional MCP capabilities (extended web research, deeper BIM inspection, premium image and video models).
If you upgrade mid-Task, the in-flight Jobs continue on their current tier, and any subsequent Jobs you trigger will use the upgraded tier — no need to restart.
Why Tasks lead to better outputs
The Agent Panel does three things that a blank prompt cannot:
- Removes the guess-work. Each Task is pre-wired to the right agent, model, MCP servers, and output shape.
- Reduces prompting effort. You describe the goal, not the prompt engineering. The orchestrator owns the prompt.
- Delivers contextually accurate, professional results. Tasks carry their own context budget — brand tokens, prior conversation, attached sources, audience profile — so the output stays on-narrative across sub-tasks and platforms.
FAQ
What is the difference between a Task and a chat prompt?
A chat prompt is a single message. A Task is a pre-wired plan: it picks the specialist agent, decomposes the work into sub-tasks, calls the right MCP servers, and shapes the output for the platform you choose. You can still chat freely if you prefer — Tasks are an opinionated shortcut, not a replacement.
Do I need to know which specialist agent to pick?
No. Each Task ships with a recommended agent. The orchestrator can also re-route mid-run if a Job is better handled by a different agent (for example, a Slides Task that needs deep BIM detail will hand a sub-task to Agent No. 5 transparently).
What are the output formats I can pick from?
Today: Instagram (caption + hashtags + 1:1 / 4:5 / 9:16 visuals), LinkedIn (post or carousel script), TikTok / Shorts (hook + script), X (Twitter) thread, PowerPoint (.pptx), Word (.docx), PDF, Markdown, and a Reveal.js web deck. New formats are added regularly.
Are MCP servers something I have to configure?
No. EXXY® AI Studio's MCP servers are managed by the platform and selected automatically per Job. You do not install, host, or authenticate against them yourself.
What if the result is not what I wanted?
You can re-run any individual Job ("redo the hook"), edit the Task plan ("add a closing CTA frame"), switch the output format ("show me this as a Word doc instead"), or fork into a free chat with the same agent. Nothing is locked.
Where do I learn about advanced Tasks?
Start with the agentic AI orchestration overview, then the deep dives on presentations, BIM analysis, and meeting assistants.
Try it
Open EXXY® AI Studio, click the Agent Panel, and pick a Task that matches what you want to publish. Choose your output format. Review the plan. Hit Run.