Create an agent with two other departments.

Creating the agent may be the easy part. Getting three departments to agree on what it should do can be harder. AI can help with both.

Bring people from your department and at least two others together around a shared problem or opportunity. Is there information that never reaches everyone in a usable form? A project stalled because the departments understand it differently? Something you could do together that none of you could do alone?

Imagine marketing, IT and customer service combining what they know about customers to see what people love and hate about your product, then developing ways to improve it. Or finance working with two other departments to find redundant expenditure without losing useful capabilities.

Start by giving each department room to explore its own needs. Then bring the contributions together, in person or asynchronously, and choose something worth building. Separate what the first version must do from nice-to-haves and dream capabilities. AI can turn the agreement into a prototype, help everyone try it, collect their responses and revise it until every department approves the same working version.

You don't need to design a new collaboration system first. The lead prompt investigates what your company already uses—Teams, Slack, shared documents or something else—and establishes a way to share prompts, results and prototypes with people and their LLMs. Products and capabilities change quickly, so it also searches for current tools where needed.

The examples below show useful directions. The company stories come from suppliers or implementation partners; the proposed exercises are ideas to adapt to your own work.

Step 1 — For the user

Use the lead prompt to get the collaboration started. It will help you find the right people, establish a simple shared workspace and organize the work through to a tested tool. Here are a few directions to explore.

Find costs hiding between departments

Microsoft describes Dow bringing supply chain, IT and operations specialists together, then piloting agents to find freight-invoice discrepancies. The account reports identified opportunities; its large savings figures were projections. A smaller exercise could connect finance, procurement and operations around one kind of expenditure.

https://www.microsoft.com/en-us/worklab/ai-impact-at-dow-copilot-identifies-millions-in-cost-savings

Give everyone a shared view of a customer

Slack describes Mural bringing account records, call summaries and support signals into channels used by Sales, Customer Success, Legal and other teams, while developing shared AI account workflows. Consider an agent that gives your own customer teams a source-linked view of the same account.

https://slack.com/customer-stories/mural-story

Connect customer insight with product development

PA Consulting reports building Unilever's DelphiAI with business and technical teams, making consumer and product information useful to marketing and R&D. For your group, a first version could bring together evidence about one product and help the departments explore improvements.

https://www.paconsulting.com/client-story/unilever-using-ai-to-empower-people-to-supercharge-innovation-and-drive-growth

Step 2 — For the user

When the lead has established the sharing arrangement, give the departmental brief and prompt to every participating department, including your own. Each group explores separately and confirms its contribution before the joint discussion. Return those contributions to the lead conversation; it will guide project selection, building and repeated review.

Done when

Your department and at least two others have tried and explicitly approved the same useful working agent or prototype. Everyone can access the agreed version, knows how to use it and knows who owns its upkeep. The requirements, tests, source references and departmental approvals are saved with it. An unresolved objection, missing department response or unbuilt capability remains open work.

Guidance for the AI

These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result.

Start with the activity and the user's context

Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear.

Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them.

Check the capabilities needed for this work

Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit.

When useful access is missing:

• Explain what you cannot reach and how that material would help.
• Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot.
• If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response.
• Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available.

Ask focused questions and respect the answers

Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful.

Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements.

Keep evidence and independent contributions distinct

Distinguish evidence, inference, and missing information. Check your work against the source material.

Use independent agents when your system actually supports delegation and independent investigation or review would improve the work:

• Give each agent a bounded brief and the context appropriate to its role.
• Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations.
• Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately.

Produce, check, and revise the result

Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them.

Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result.

Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized.

Your AI prompt

Help three departments build something useful together

Help me bring my department and at least two others together to create an AI agent that addresses a shared problem or opportunity. The agent may be a shared assistant that answers questions from agreed sources, a tool that assembles information, or a workflow that takes agreed actions. Help us choose what is useful, build a working version, gather each department's responses, and improve it until every participating department explicitly approves that version.

I may have only a vague idea. Carry the research, documentation, synthesis and building work; ask me focused questions one at a time where people, access or decisions are needed. Start by establishing my department and any partners or shared issue I already have in mind. Do not select the project for the other departments.

Find the people, existing work and a way to collaborate

Inspect the relevant connections you can actually use. Search accessible email, Teams or Slack messages, other relevant chats, calendars, meeting transcripts, project documents, directories and existing AI-tool inventories. Use the people, issue or department names already available to bound the search. Look for cross-team handoffs, repeated questions, stalled decisions, duplicated effort, shared opportunities and any similar project already underway. Use invitations, roles and recorded decisions to identify likely participants, their actual concerns and who can approve a shared trial. Cite findings and distinguish them from your inferences.

Help me enlist a named representative from each participating department, including mine. Have the people confirm who can approve for their department and how they will bring relevant colleagues' needs into the work. Do not treat an inferred organizational role as delegated authority.

Find the company's existing collaboration and AI setup before proposing a new service: Teams and SharePoint, Slack and shared documents, Google Workspace, an existing project system, or another approved arrangement. Search internal guidance on sharing company information, agent creation and available licenses. Check current official documentation and prices for any product-dependent recommendation. This field changes quickly; do not rely on remembered feature names or assume everyone has the same paid plan.

Choose a simple arrangement with me that specifies:

• Where the common project brief, prompts, departmental contributions and current source references live.
• How a person and their chosen LLM can actually read the material; verify a representative file or link rather than assuming access.
• Where the current prototype, numbered versions, feedback and decisions live.
• Who coordinates updates and how each department returns a response.

Use the company's existing system wherever practical. Teams can be a home for collaboration and, where supported, a shared agent; Slack can hold discussions and structured returns. A shared folder with ordinary documents or Markdown files is also a valid starting point. The communication channel, the files and the agent need not be the same product.

If a connection is missing, investigate the supported setup first and ask for the smallest necessary access action. If direct sharing with an LLM is unavailable, prepare clearly named exportable files for people to paste or upload into separate sessions. Explain which materials must stay within each department; sharing a summary must not imply permission to reveal its underlying records. Have the contributor confirm what can enter the shared packet. Preserve links or source identifiers for authorized verification.

Create the workspace or files using available tools within my authorization. If a company administrator must do a step, prepare the exact request and continue work that does not depend on it. Draft invitations and messages; send only when I explicitly authorize sending. Check that all representatives can use the agreed return path before relying on it.

Let each department explore independently

Give every department, including mine, the same neutral brief and a self-contained exploration prompt. If I supply the separate departmental prompt from this activity, use it. Otherwise create a prompt that interviews that department, searches its relevant records, and produces two to four candidate opportunities, with evidence, desired outcomes, likely partners, requirements, nice-to-haves, dream capabilities, feasible first versions and unresolved questions. Ask each department to review and authorize its own contribution for sharing.

Collect these initial contributions separately before showing the departments a preferred solution or synthesizing their positions. Set an agreed return date with the people involved. Keep original contributions intact with department, representative, date and version. Do not invent an absent department's response. If something is missing, identify the gap and help obtain it before making a shared choice.

Choose the shared project together

Read the contributions and their supporting evidence. Group related ideas while retaining disagreements, different definitions and the value each department wants. Identify work that actually needs the departments together. If useful, research a few relevant real examples with primary links, distinguishing supplier claims from independently established outcomes.

Possible starting points include marketing, IT and customer service combining customer evidence to identify what people love and hate; finance with procurement and operations finding redundant expenditure while preserving useful capabilities; or teams resolving a project stalled by missing information or conflicting interpretations. These are proposals to explore, not assumptions about our company.

Prepare a short, source-linked comparison of promising projects: who benefits, what each department contributes, what is currently missing, and what a useful first version could do. Facilitate either an in-person/remote conversation or asynchronous comments in the agreed workspace. Use actual contributions to frame clarifying questions. Record the chosen purpose, participating departments and explicit agreement to the first scope. If they disagree, expose the concrete tradeoff and help them resolve it or choose another project; do not manufacture consensus.

Turn the agreement into a buildable specification

Consolidate the requirements into one current document while preserving each department's original words and sources. Separate:

• Requirements: capabilities needed for the agreed first version to be useful.
• Nice-to-haves: useful additions that can wait.
• Dream capabilities: the larger ambition to preserve for later.

For every requirement, identify the department that needs it and an observable example of success. Agree on the intended users, inputs, source access, outputs, permitted agent actions, and when a person must decide. Define the actual first version in plain language. A shared assistant with a reliable source set may suffice; another project may need a small app or connected workflow.

Create a compact set of realistic test tasks with the departments, including one contribution or handoff across department boundaries. Establish the expected result from records or human judgment before using it to assess the agent. Include relevant missing-data or conflicting-source cases. Avoid measures that simply prove the software ran.

If the proposed scope is too large, suggest a smaller slice that still performs a complete useful job. Get explicit departmental agreement on that slice and the tests before building dependent functionality. Preserve the deferred ideas. Document actual tool choices, costs, access dependencies and a build plan suited to the available AI and development environment.

Build and test an actual prototype

Use the available agent builder, coding tools or other approved environment to create the working agent. If a different AI tool is needed, investigate the current route and prepare a complete handoff containing the specification, approved source references, decisions, test examples, permitted actions and remaining work. Help carry the build through there. A specification or a link to an agent builder does not complete the activity.

Build one prototype, or two meaningfully different versions if comparison will settle a real question. Keep the versions identifiable. If real integration is unavailable, label sample data and simulated actions, and distinguish what the prototype demonstrates from what remains unconnected. Do not quietly substitute a mockup for the agreed functioning agent.

Divide large source review or development into bounded assignments when useful. Each AI worker receives its question, permitted source set, output and completion test. Track coverage and reconcile conflicting findings. Use separate review of the actual prototype and evidence where available; if agents are unavailable, use manageable passes or fresh sessions with complete handoffs. None of these AI roles can represent a real department's approval.

Run the agreed tests, inspect the outputs against the evidence and correct failures. Show what the agent actually did, including any incomplete connection or unsupported answer. Deliver a usable prototype link or artifact and brief instructions for trying it.

Return it to every department and repeat

Prepare a review packet with the exact prototype version, agreed purpose and requirements, a few tasks to try, known gaps, and an easy return format: department and reviewer, task attempted, expected versus actual result, change requested, and whether the version is approved. Distribute through the agreed route when authorized, or give me the ready-to-send packet. Confirm access rather than assuming a shared link works.

Ask each department to try it on work it understands. Collect its actual response and supporting examples, including any additional colleagues the representative needs to consult. Keep a response table for every participating department: awaiting response, changes requested, or explicitly approved, tied to a version and date. Silence, attendance and an AI prediction of acceptance do not count as approval.

Bring the responses back into the lead session. Reconcile conflicting requests with the people affected, revise the specification when needed, build the changes and rerun the relevant tests. Share the revised version and repeat. If a later change affects a department's agreed behavior, obtain its approval of that version; do not carry an obsolete approval forward as if nothing changed. Distinguish a request beyond the agreed scope from a failure to meet that scope and ask the departments how to handle it.

Continue until every participating department explicitly approves the same usable version. If disagreement, missing access or a missing response prevents completion, keep the project open with the precise unresolved item and the next human decision. Do not wear people down into approval or declare completion because several rounds have elapsed.

Leave a tool people can keep using

Save the approved agent or prototype and its source/configuration, agreed requirements, tests, source references, short usage instructions, change history and version-specific approvals in the chosen company location. Confirm who owns upkeep and how people report a problem or propose a change. Make any further production integration or rollout a visible next decision when it lies outside the approved scope.

End with what each department can now do using the working version, the evidence from its trials, and any remaining limits. A plan, a promising demo, or approval from only two of three departments is an interim result.

For the colleague

We are exploring an AI project with our department and at least two others. Please use this prompt with your chosen company-approved LLM to work out what would be useful from your perspective. Begin independently, before reading the other departments' proposed solutions. Review the result and confirm what can be shared. The coordinator will provide the agreed return location and date.

Your AI prompt

Find a shared opportunity from our department's perspective

Help our department identify useful AI projects to explore with at least two other departments. We are contributing independently before the group selects a project. We may already have a neutral project brief or a return format; use it if supplied. We do not need a finished idea.

Start by finding out which department we are, what I do and what I hope could work better. Ask one focused question at a time, building on anything I provide. Welcome rough notes and examples.

Investigate our actual work

Inspect the company sources you can actually access. Search relevant emails and chats, meeting transcripts, project documents, process records, requests, support tickets, calendars and handoff information using the department, people and topics already identified. Look for recurring friction with other teams, information that arrives too late or in an unusable form, questions nobody can answer alone, stalled decisions, duplicated effort, and opportunities that require combining departments' knowledge.

Find existing projects or tools that might already address the issue. Use roles, correspondence and documented concerns to identify the people affected and the other departments whose involvement matters. Distinguish their recorded views from your inferences; do not speak on their behalf.

If access is missing, investigate available connections or a permitted export before asking me to reconstruct the history. Work with specific examples I can supply when necessary, and label those as an incomplete account. Keep the investigation proportionate; do not require a department-wide data collection exercise just to suggest a useful first project.

Explore possibilities with us

Discuss what you find, ask about what matters most, and offer two to four distinct candidate projects grounded in our work. Consider both problems and new opportunities. Possibilities include joining customer insights across marketing, IT and service; examining duplicated expenditure with finance and two operating teams; making an awkward handoff clearer; or surfacing the missing decision that keeps a project stalled. Treat these as inspiration, not a prescribed answer.

For each candidate, explain the current situation with a concrete example and source, the outcome we want, why other departments are necessary, what our department could contribute, and what a small working agent might actually do. Identify unknowns rather than promising that a connection or capability exists.

For our preferred ideas, separate requirements, nice-to-haves and dream capabilities. Suggest one realistic task that would demonstrate usefulness and what a satisfactory result would look like. Preserve differences of opinion within our department. Help me involve a relevant colleague if essential knowledge is missing; prepare the questions or message rather than claiming you consulted them.

Prepare our contribution for the other departments

Create a concise, shareable contribution with our department, named representative, date and version; candidate ideas and preferences; evidence and source references; desired outcomes; requirements, nice-to-haves and dreams; proposed first versions and test examples; likely partners and questions for them. Keep our initial contribution independent of other departments' proposals until the lead has collected everyone's first pass.

Separate material permitted for the shared group from restricted details. Summarize only what we are allowed to share, and preserve a way for an authorized person to verify the supporting evidence. Ask us to correct the contribution and explicitly confirm what may be shared. Mark ideas as proposals, not decisions by the other teams or approval of a future tool.

Save the reviewed contribution in the requested format or produce a clean file ready to return to the coordinator. If a shared workspace is specified, check that you can actually use it. Send or post to others only when explicitly authorized. Finish with our confirmed contribution and the open questions we want the joint group to explore; do not begin building the shared agent or infer group agreement.

What to send back

Return your reviewed contribution: department and representative, date and version, candidate ideas and preferences, supporting evidence that may be shared, requirements, nice-to-haves, dream capabilities, a useful first-version test and questions for the other departments. Say what remains uncertain. This is an exploration contribution, not approval of a project or a finished agent.

These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.