Test one new way of working every week.

What the AI will do

• Learns how your team works from emails, messages, calendars, meetings and project records, then looks for small ways to reorganize that work around AI.
• Builds a recurring experiment companion: it might replace an update meeting, coordinate work divided in new ways, or handle and collect interruptions until a better moment.
• Helps run the experiments you choose and uses their results to improve what it suggests next.

What you can provide

• A team or workflow to start with. Existing access helps; rough notes, messages or a calendar export can also get things moving.

Where you'll take part

• Choose a small change worth trying and settle how it will affect colleagues' work.
• Choose the cadence, judge whether each experiment helped and decide what to keep, change or stop.

What you'll get

A working cycle for suggesting and trying small changes, with a growing record of what your team learns. The AI shows what is running, what needs a missing connection and which trial results are still to come.

Activity context

At Feedforward, we keep coming back to an idea: getting the most from AI means reorganizing how work happens around what it can do. You can explore that in small pieces. A meeting becomes a useful exchange of updates without everyone gathering. A project can be divided differently because AI keeps track of the moving parts. The interruptions that break your concentration get answered or collected for a better moment. Over time, little experiments can reveal a very different way for your team to work.

For the user

Name the team or a frustrating piece of work. The AI investigates what happens, helps you choose a small experiment and builds what it needs. You bring the judgment about what is worth changing; the system keeps the history so each trial teaches you something for the next.

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 my team discover what becomes possible when we reorganize work around AI's capabilities. Build a recurring tool that learns how our work actually happens, proposes small useful experiments, helps carry out the ones we choose and accumulates evidence about what works. Keep this grounded in our work, with the AI doing the investigation, coordination and building.

Learn how the team gets work done

Start with the team, project or rough description I supplied. Inspect and try the available connections. Search team emails and messages, including WhatsApp where accessible, calendars, meeting transcripts, project plans, task histories, shared documents and completed work. Use names, project terms, deadlines and recurring meeting titles to follow whole episodes: what started a piece of work, who contributed, what was handed over, what interrupted it and what useful result emerged.

Look for recurring information exchanges, decisions, status meetings, requests for help, rework, waiting, coordination problems and useful practices worth extending. Find existing AI experiments and tools so suggestions build on what we have tried. Establish the people involved, their actual roles and the concerns they have expressed in correspondence and meetings. A busy calendar does not prove wasted time, and an interrupting message may be essential work; examine its purpose and aftermath.

Keep sources, dates and coverage visible. For a large team, assign bounded workflow or project histories to agents with explicit sources, questions and deliverables. Keep connected handoffs together, use successive waves when needed and reconcile their findings into one account. With no agents, use manageable passes and a saved coverage record.

If access is missing, investigate permitted connections or exports before asking for the particular action only I can take. Rough notes, a few message threads and a calendar export can start the work. Explain which patterns remain hypotheses when the record is thin. Ask one useful question at a time about an uncertainty that changes the experiment; continue unaffected investigation rather than requiring a prepared workflow document.

Find small experiments that change the organization of work

Show a short account of how work currently happens and offer a few concrete experiments grounded in the record. Each should change who or what does a piece of work, how people coordinate, or when information reaches them. Explore possibilities such as these when they fit:

• Replace an information-sharing meeting with an AI tool that gathers the updates, resolves straightforward questions and gives each participant what they need. Find the meeting's other functions first: decisions, disagreement and relationship-building may need a different home.
• Parcel out a project in a way that would once have created an unmanageable coordination burden. Let AI maintain the task and dependency record, prepare handoffs, track missing inputs and reconcile the pieces, while people provide judgment where it matters. This may involve colleagues, AI agents or both.
• Investigate recurring interruptions. Have AI answer or complete the ones it can handle within agreed authority. For others, create a parking lot that retains the original request, requester, urgency and deadline, then brings requests back at useful moments between focused work or project deadlines. Requests with genuine time sensitivity need an agreed route through; a parking lot should not make work disappear.

These are examples, not a required sequence or the limit of the search. Look for other ways AI's current capabilities could make a better arrangement feasible. Read current official documentation and try a small task before relying on a capability. Distinguish a functioning tool, a plausible experiment and a speculative possibility.

For each proposed experiment, describe the current arrangement, the specific change, the hoped-for benefit, the people affected, the AI work, the human work and a small observable result. Use real team concerns to judge whether a change would help. Let me choose an experiment or supply another. Confirm material commitments involving colleagues with the responsible person through the appropriate authorized route; do not infer their agreement from my interest in an idea.

Make the chosen experiment runnable

Bound the first trial to a meaningful workflow and period. Establish its present performance from the record where possible: useful output, elapsed time, waiting, human effort, errors, interruptions or whether recipients actually got what they needed. Choose measures that fit this task; do not equate more AI usage with improvement. If the baseline must be observed first, set up that observation and label estimates separately.

Write a compact experiment brief: the hypothesis, exact change, work and participants covered, expected output, what will be measured, where the result will be recorded and the review date. Include the effort of checking, correcting and maintaining the AI work. Note other changes during the trial that could explain the outcome. Choose a sensible comparison, such as recent similar work or a comparable unchanged workflow; explain what that comparison can and cannot establish.

Build the selected tool and its working materials using the team's available platforms. Carry out the actual creation of prompts, source connections, request queues, agent briefs, shared project records or information summaries that the experiment requires. For project coordination, preserve dependencies, owners and completion evidence rather than merely creating more tasks. For meeting replacement, produce the information people would otherwise need the meeting to obtain. For interruptions, implement the answer-or-park decision, visible queue and timed release with the agreed urgent exceptions.

Inspect runtime, storage, scheduling, integration and delivery capabilities. Explain the specific prerequisites the chosen experiment needs and prepare any user-only setup action. Test representative ordinary cases and relevant exceptions in an appropriate test space. Verify actual handoffs and outputs, not just the appearance of a screen. Before activating a change to colleagues' work or sending messages, settle the concrete scope and standing permissions; once authorized, carry out the agreed setup without seeking approval for each covered routine step.

Build the recurring experiment companion

Create a persistent record of candidate experiments, sources, chosen changes, settings, baseline, dates, observations, outcomes and decisions. Record abandoned or disappointing trials as well as successful ones. The next suggestion should learn from these results instead of proposing the same thing under a new name.

Set up a regular review of the connected work sources and this experiment record. Weekly is a useful starting cadence; let me choose what fits the team. New suggestions should explain what changed or what earlier trial they build on. Present one manageable experiment at a time unless we choose a wider effort. Keep ideas awaiting attention in a backlog and avoid interrupting an unfinished experiment without a concrete reason.

Configure the source refresh, schedule, durable instructions and destination for suggestions and results. Verify that the tool can run after this conversation ends, retain its history and identify failed or stale source reads. Test one scheduled or triggered cycle end to end, including a new source item, an updated suggestion and the agreed delivery destination. An authorized test notification must be clearly recognizable as a test. Provide a way to pause the system, correct its understanding and change the cadence.

If an essential connection or scheduler cannot be enabled, deliver the working available pieces, the reusable invocation and retained experiment record, identifying automatic operation as unfinished. Continue the first experiment with the agreed manual bridge where useful. Do not call a list of ideas a working recurring tool.

Run the trial and learn from what happened

Start the chosen experiment when its necessary agreement and access are in place. Track the actual work, exceptions, checking effort and outcomes. Where a trial takes days, arrange the observation and follow-up so it can continue beyond this session; give me the first verified outputs and the future review date without inventing results that have not happened yet.

At review, compare the observations with the baseline and intended result. Ask the people involved for the focused judgment the record cannot supply, using authorized outreach or a prepared question for the user to collect. Did the change help them do useful work? Did it merely move effort to someone else? Preserve their feedback and conflicting evidence.

For a substantial experiment, give a separate reviewer the brief, real outputs and observations without the builder's explanation of why it worked. Ask it to test fidelity to the original purpose, missing effort and alternative explanations. Use a fresh session or a clearly identified shared-context checking pass if independent review is unavailable. Repair the tool where needed and recheck affected behavior.

Discuss whether to keep, change, repeat or stop the experiment. Record the choice as ours only after it is made. Explain the conditions under which a promising result might transfer to another workflow; one successful trial is evidence about its setting, not every team. Use this learning to make the next experiment more useful.

Deliver a working cycle and an accumulating history

Give me the actual experiment companion, its source and scheduling status, the first runnable or active experiment, the materials it uses, and the place where results and decisions accumulate. Include a short demonstration of the next suggestion and how an earlier result influences it. State which initial checks passed and which real-world results are still pending.

The aim is a growing understanding of how AI could transform this team's work through small changes people can actually try. Keep the continuing scope to the sources, cadence and authority we agreed, with a clear way to pause or revise it. A larger reorganization can emerge from what we learn; do not represent one as already chosen.

Done when

You have the recurring experiment companion, a first runnable or active trial and a place where its measures, observations and your decisions accumulate. Source updates and scheduling have been tested where available; any missing automation and real-world results still awaiting the trial are explicit.

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.