Imagine organizing a weekend workshop for 50 people. Participants should bring their own projects and leave with something small they can demonstrate. The budget is limited, and the venue is undecided.
An agent finds several spaces, compares prices, and produces an attractive schedule. Then the plan stalls. Affordable venues lack equipment; equipped venues cost too much. Another keyword search might uncover something missed, or simply return the same packages again.
An experienced organizer may try a different move. They might ask a venue to provide equipment for a public open day. They might invite an instructor who already owns the equipment, reducing the demands on the space. Or they might run a small trial and discover the workshop needs less equipment than expected.
These actions change what can be arranged next. ToWow uses PFE—the Possibility Formation Engine—to investigate such choices: when the current plan cannot work, which change is most worth making next?
We use the workshop as a design example to explore how an agent could organize that process and make visible what each attempt actually contributes.
A query cannot add a power socket
If the missing fact is whether a venue is free on Saturday, a query can move the plan forward.
If the venue lacks a suitable electrical connection, asking ten times will not add a socket. What is needed is an equipment arrangement, a site inspection, or another location.
If the instructor worries that 50 beginners working simultaneously will be unmanageable, a rotation between small groups or a rehearsal with assistants may help. The problem concerns how the collaboration is organized.
If the organizer cannot yet say what participants should make, requesting a precise venue quote may be premature. Two activity prototypes could be more useful: one small piece per person, or a more complex project per group of five? Seeing the alternatives may help the organizer decide.
“We don’t know” therefore needs an object: the timing, equipment compatibility, the organizational method, or the experience we prefer. That object determines whether to read, whom to ask, what to try, or what to redesign.
PFE keeps this diagnosis in the system. A model can propose it, but it can also come from a real error, a participant’s answer, or a tool check. New information should revise the diagnosis. Its purpose is to help select an action.
Unpacking “publicity in exchange for a venue”
The organizer proposes a new arrangement: make the workshop a public open day, bringing potential customers to the venue in exchange for equipment access.
The proposal changes a transaction previously compared only by rent into a combination of rent, equipment, activity, and promotion. It may be worth discussing. It may also turn a simple rental into a harder exchange. The venue might want long-term tenants and place little value on brief publicity.
The system cannot now enter “free equipment” in the budget. It needs a concrete invitation: which equipment, at what times, for what use, and what the event could offer the venue. The response will show whether this path actually adds usable conditions.
Several actions can address the same blockage:
| Action to try | Intended change | A result that would alter the plan |
|---|---|---|
| Ask about availability and equipment | Turn a vague quote into a specific usable arrangement | Only half of Saturday is available; shorten the activity or change the date |
| Have the instructor trial a small group | Observe equipment, pacing, and guidance needs | Groups of six still need frequent help; add assistants |
| Invite a partner with equipment | Add equipment and guidance together | The partner can provide equipment but only attend in the evening |
| Build a shared project starter kit | Reduce installation and explanation time on site | Beginners can check their own environments, freeing up guidance time |
| Compare two activity prototypes | Help the organizer develop more specific preferences | Conversation matters most, changing the weights given to venue and duration |
People, existing services, or tools can perform these actions. The agent connects the present difficulty with a suitable action, prepares the materials, and uses the result to revise the plan.
A failed conversation with the venue need not lead to another wording of the same request. Other actions may change the equipment requirement itself or bring in a partner who can supply it.
Make room for a real trial
Suppose the organizer’s biggest concern is whether participants can produce something in two hours. Spending three days refining the full schedule may be less useful than inviting two intended participants to try one part of it.
The trial can be small: supply starter materials, ask them to attempt the critical step, and record where they stop, what help they need, and what they produce.
If both get stuck installing their environments, build a preconfigured environment. If installation works but the task is unclear, revise the explanation and project size. If they finish quickly, some of the teaching time may be better used for conversation.
Each result implies a different change. The value of the trial lies in the choices it changes.
Design practice has long used this back-and-forth. The Design Council’s Double Diamond places understanding the problem, redefining it, developing alternatives, and small-scale testing within one design process. PFE asks how, when agents perform part of that work, observations, actions, participant responses, and plan revisions can remain connected and available to the next action.
Write the changes back into the plan
Ordinary data structures are a reasonable starting point for software. Keep the current objective, known conditions, unresolved questions, and several candidate actions. Attach an expected effect, required resources, and a way to inspect the result to each action.
The system might retain two competing judgments: “installation is the main obstacle” and “understanding the task is the main obstacle.” They support different work: preconfiguration and course revision. After the trial, record which interpretation the observations support and whether that changes the chosen action.
To let the next executor continue, record the before-and-after difference. For example: “We previously budgeted one device per person. The trial showed that three people taking turns could achieve the objective. Device demand fell, but we need rotation instructions.” That is more useful for planning than “trial successful.”
Authority follows the particular action too. An agent can organize materials and build a starter kit within existing authorization. Venue bookings, spending, and commitments to outside partners go to the relevant decision-maker. The interface should present a question where a decision is actually needed, accompanied by the comparison materials already prepared.
The cost of the next step includes other people's time
It is easy to keep extending an action list. A working system has to choose.
A question takes an instructor’s attention. A trial requires people to attend. Building a tool may take more development time than the workshop itself. A worthwhile action should have a chance of resolving a consequential uncertainty or creating conditions valuable enough to repay those costs.
Three questions are a useful starting point: which choice would this action change, how long until its result arrives, and which paths remain if it disappoints?
There is no need to collapse everything into a score immediately. Early on, showing the alternatives and reasons side by side may help the organizer more than an uncalibrated “optimal action: 0.87.” As real records accumulate, we can investigate which signals predict useful outcomes.
Stopping is also a concrete choice. If coordinating a suitable venue already exceeds the budget, an established provider’s standard workshop may be preferable, or the event may wait until partners and resources are available. The decision should explain which value is preserved and which is deferred.
Let ordinary workflows take over when the arrangement matures
The first workshop may need repeated trials, equipment coordination, and revisions. After several events, the instructor has a starter kit, the venue has clear availability conditions, and the organizer knows which scale works.
Those lessons can become reusable materials, booking rules, and routine processes. The next workshop can use them directly, reopening only the parts that need to change.
PFE can therefore sit inside existing project tools or agent workflows. Existing orchestration handles stable steps; a formation strategy intervenes when conditions are missing or new situations arise. Its added value must ultimately be compared on the same tasks with mature workflows, strong models, or an experienced organizer.
At the next event, the tried starter kit and ordinary bookings may handle most of the work. Exploration resumes only when objectives or conditions change. PFE’s value should appear across this cycle: progress when conditions need to be created, and a smooth handover to ordinary work when they become stable.
Materials: rewritten from the diagnosis of unknowns, formation actions, counterexample-driven loops, and reuse of stable relationships in ToWow’s historical PFE designs. The workshop is a teaching construction, not a customer case. Exact research locations are recorded in the publication package’s SOURCE-NOTES.md.