The Folder
You Own

Ask someone what they did in their last job and you may get a title, a department and a list of duties. Ask what other people relied on them to notice, and the conversation changes. There was the request that always arrived incomplete. The customer everyone thought was being difficult. The handover that looked finished until you asked who would make the next call.

Those adjustments were part of doing the job properly. They rarely acquired names. When the job disappears, the title is easy to put on a CV. The judgement that made the person useful can be much harder to bring with them.

That is where I want Songbird to begin. Before asking someone to become a founder, join a venture or describe their unique contribution, help them see what they have already been doing. Put some of their own evidence on a table. Work through it beside them. Write an account they can correct, recognise and take away.

This book describes the first gift I want Songbird to make. We have experience helping people retell an idea they bring us. We do not yet have a completed, hand-made personal folder that demonstrates the whole service. The visitors and specimen pages ahead are labelled illustrations; the practical protocols describe work still to do. That boundary matters because a persuasive description of a service can otherwise begin to impersonate its results.

Competence hides its own workings

Michael Polanyi's starting point in The Tacit Dimension is that knowing can exceed our ability to explain. He uses recognising a face: we can recognise the whole without being able to specify all the features by which we know it. That is a useful starting point for this problem, rather than proof that every skilled person is incapable of describing their skill.¹

At work, attention goes towards the problem. Once an adjustment becomes familiar, the person may stop treating it as an adjustment at all. They ask a question early because experience has taught them what happens when nobody asks it. Afterwards, they remember resolving the request. They may leave out the question that prevented a much larger mess.

An inventory of duties misses this. It records what the job required, while discovery needs to recover how the person recognised the situation and what they did with that recognition. A story about one awkward handover can contain more of that judgement than a page of confident adjectives.

Familiar to you, useful somewhere else

The second difficulty sits outside the person. You know the place where you learnt something. You may know much less about the places where that knowledge is missing.

A nurse knows what seems obvious among experienced colleagues. That does not tell her which observations would be unfamiliar to a team designing a service for families. Equally, being unfamiliar to that team would not establish that her knowledge is rare across a profession, or that anyone will pay for it. Those are separate questions requiring comparison.

The outside eye has work on both sides of the table: understand the person's evidence, then understand a real need well enough to propose a connection. Calling everything transferable is no help. A useful account says what may transfer, to whom, under what conditions, and what remains untested. We will put that demand to work in the nurse's folder.

The story you already tell

A title is also a way of arranging memory. Once you think of yourself as the person who manages a team, you may interpret every good result as evidence of management. The patient explanation, the unusual question and the well-made handover all become supporting detail. Someone else might see a different pattern.

The Johari window offers a simple language for differences between what we know about ourselves and what others notice. Joseph Luft's account includes a blind area: things apparent to others that are not apparent to us. It also cautions against forcing awareness. The model is a way to discuss a relationship, not a licence for an observer to announce someone's hidden identity.²

An observer can be wrong. They bring their own favourite explanations, status assumptions and commercial interests. A company looking for founders can start seeing a founder in everybody. A discovery partner earns trust by making the interpretation visible enough to dispute.

Give memory something to work with

A blank question asks you to retrieve the answer and invent the vocabulary at the same time. A document, a situation or a proposed description supplies a cue. Raluca Budiu's explanation of recognition and recall in interface design describes why visible cues can reduce the burden of retrieving information from memory. The same distinction helps us choose a gentler starting point here, without assuming recognition is infallible.³

We can ask whether a particular description fits. The answer may be an immediate refusal followed by a precise reason. That reason is useful material. It is also possible to agree with a flattering description because it feels good. Recognition begins an investigation; it does not finish one.

Self-promotion must not become the entry fee to discovery.

The person with a long career has more material to choose from. The new graduate may bring a group project, a part-time job or something they made for a friend. The starting evidence differs. The care owed does not.

And the work need not reveal a business. It may produce language for a conversation, a clearer boundary around a role, or a better account of an experience someone had dismissed. Those can be worthwhile things to leave with.

The next decision is who helps make that account, and how much authority we give them.

The helper's authority begins with showing their working. They can propose a connection, bring a comparison and write a better sentence. They cannot turn a tentative reading into a verdict about who you are.

I want someone working beside you, doing the writing while you work out whether it is true. That division of labour is the service. The visitor should not have to arrive with a polished account of their capabilities and then pay, in attention or disclosure, for somebody to reformat it.

I call this person a discovery partner. An assessor has a selection to make. A personality test has a classification to offer. A coach may help someone pursue a goal they can already name. Those activities can be useful, but the task here is earlier: find and articulate something the person may not yet know how to ask for.

Who does which work?

The partner prepares material worth reacting to. They listen for distinctions: what the person noticed, what they changed, what they enjoyed, what they would never want to do again. They separate a reported incident from an interpretation of it. Where they propose value in another setting, they make the comparison explicit.

The visitor contributes what they choose. They can tell a story, bring a document, point to an inaccurate sentence or say that the question is going somewhere they do not want to go. They decide which descriptions fit and which pages may leave the table. Agreement with an interpretation does not automatically authorise sharing it.

There is real work on both sides. A person may need to pause, remember an uncomfortable detail or correct a flattering mistake. The partner makes that work easier by doing the preparation and writing, rather than returning the hardest question in a larger box.

None of these is a required upload. A document is something brought to a meeting. If the visitor cannot appropriately share it, a recreated, non-confidential example may be enough to discuss the underlying judgement. Access to a file is not the same as permission to contribute it.

The idea points back towards its holder

Suppose someone arrives wanting to build a tool that prevents customers from repeating their story to several departments. It is tempting to begin designing the tool. First, ask what happened.

Perhaps they watched three handovers fail because every department believed another one owned the next step. Perhaps they were the person customers called when that happened. Perhaps the idea came from reading an article, and they have never handled the situation themselves. These are different starting points. The same proposed product can conceal very different relationships to the problem.

Following the idea back to an incident gives us something to examine about the person. It does not prove the product is viable. It may reveal that their useful contribution is explaining where responsibility becomes ambiguous, even if their proposed solution is poor. We can preserve that observation without requiring them to defend the entire business idea.

Someone who arrives without an idea is equally welcome. Discovery should not depend on already knowing the language in which an opportunity will eventually be described. Search is useful once you know what you are looking for. This work often has to help create that knowledge.

Different beginnings, equal attention

Consider two illustrative visitors: a fund chief executive and a new graduate. The executive may bring a detailed understanding of an industry and disagree with the partner's comparison immediately. The useful question could be which decision looks routine to them but remains difficult for less experienced colleagues. Their position does not entitle the service to confidential institutional documents.

The graduate has less career history. They may bring a project where they discovered that the group had interpreted the brief in incompatible ways. Ask how they noticed, what they changed and what happened next. One incident will not establish a mature capability. It can still provide a specific beginning, with an honest statement of how much evidence supports it.

Adaptive pathways means the work follows the material and the person. It should never become a prestige ladder in which the executive receives thoughtful interpretation and the graduate receives a standard form.

A refusal contains a distinction

If someone rejects a proposed company page, the partner needs to find out what was rejected. The problem may interest them while the suggested job does not. The job may fit while the timing is impossible. A family commitment may make this month unavailable without changing their interest at all.

Collapsing these answers into a low-interest score throws away the useful part. Write the distinction in words the person accepts. Then change the relevant page. A refusal is allowed to remain a refusal; it need not become material for a more effective persuasion attempt.

Luft's discussion of the Johari window links increasing awareness to the conditions of the relationship and warns that forced awareness is undesirable. That is a useful restraint here. The partner should invite a contribution, explain its purpose and accept a boundary without trying to diagnose it.²

A name must earn its place

Sadie is a working persona name for the AI discovery partner. If we use it, the visitor must know they are working with AI. A name becomes useful when it stands for continuity: the correction remembered, the promised page delivered, the question not asked all over again. Simulated intimacy is a poor substitute for any of those things.

The relationship I want the visitor to experience is simple: this is our working understanding, and I get to change it. That feeling requires a visible understanding. If all the interpretation stays inside the helper's head, the person has little to inspect except the tone of the conversation.

So put the material between them. Let us see what a session actually makes.

There are three things on the table: a CV, a short presentation and a blank sheet. The visitor chooses the presentation. It gets closer to what she actually did than the list of responsibilities on her CV.

She is a fictional former customer-service manager whose role has been made redundant. Everything in this session, including the dialogue and specimen pages, is an illustration. Her presentation is a recreated example without employer or customer information. We are going to follow it all the way into a folder.

The feeling I want is a kitchen table with a whodunit box opened across it. There are clues, competing readings and something worth figuring out together. The visitor is a co-detective. She knows things the partner cannot know; the partner can put a possible pattern in front of her. Neither has to pretend the answer arrived complete.

Begin with the thing she brought

The presentation has three simple slides. The first shows an unclear request. The second shows two departments, each expecting the other to call the customer. The third replaces that ambiguity with a named owner and a note saying what will happen next.

The partner asks what made her realise the second slide was the problem. She remembers a customer explaining the same situation again. The departments already had the information. What they lacked was an agreed responsibility for acting on it.

That distinction is the first useful finding. A CV might reveal experience. A work sample might reveal a method. A story might reveal judgement. Here, the presentation shows a method, while the remembered incident explains why she chose it. The partner writes a source note and asks her to correct it.

Notice what the note does not award her: exceptional leadership, increased revenue, a proven method for every company. There is enough here to continue without adding those claims. She accepts the account of the incident. That acceptance lets the next page refer to an agreed starting point.

The first page gets the role wrong

The partner prepares a possible role: explain a small service to customers, manage the relationship and run a support team. It is plausible given her previous title. It is also wrong in a way that matters.

She points to managing the team. She does not want the rosters, supervision and recurring responsibility for everyone else's availability. Asked about the customer work, she makes a second distinction: she enjoys understanding a genuine need and explaining a useful option. She does not want to cold-call people or pressure them towards a purchase.

The partner now has something better than a sales label. They write two columns. Under Enjoys: understanding a customer's situation; explaining suitable options; making the next step clear. Under Wants to avoid: cold approach; pressure to buy; staff management; open-ended support responsibility.

One sentence initially says she enjoys making clear promises. She changes promises to explanations. She can explain the service accurately; she does not want authority to commit somebody else to delivery. That small edit changes the responsibility picture. It belongs in the record.

A less careful service could register that she likes customers and keep the salesperson tag. Then every later suggestion would inherit the mistake. Here, show–discuss–revise produces a changed page. The change is the evidence that the conversation counted.

What would a morning contain?

The revised role is still abstract, so the partner makes a possible morning. She reads an agreed brief before a booked conversation. She asks about the customer's situation and explains the relevant options. Afterwards she writes a short handover with the remaining question and the next responsible person. Her part ends when that person accepts the handover.

She likes the shape, then asks the obvious question: who follows through? The page had quietly assumed another person existed. The partner marks that responsibility unallocated. Until someone takes it, the role is incomplete.

This small company page makes the proposition tangible. It has earned its place as an artefact on the table. It has not earned an operating plan, a forecast or a decision to launch.

She says she could not consider it this month. The note records a timing constraint. It does not downgrade her interest. Asked what she would need before thinking further, she says she would have to talk it through at home.

A page made for this moment

Giving someone a standard profile with their name on it is an ordinary service. The page here has been prepared because of the distinction this person made: she likes explaining options but does not want to own every consequence of the conversation. It should feel as considered as a useful letter laid in front of her.

Venture World, my own project, already uses desk-post imagery: an opened letter and its envelope, a clipping with traces of print behind it. That gives us a visual vocabulary for something brought to the table. The point survives on a phone. It depends on recognisable pieces of work, not a simulation of a large wooden desk.

The visitor should be able to use this with headphones on, sitting on a couch beside someone else. In the intended service, the AI may speak and show material while she replies entirely through taps and typing. Speaking aloud is optional. A requirement to answer by voice would turn an ordinary shared room into a barrier to participation.

Two clocks, one honest conversation

The small agreed note can change while they talk. A substantial comparison or a carefully prepared page may need longer. We should make that difference comfortable instead of pretending every thoughtful addition was conceived instantly.

The Talker–Reasoner proposal by Christakopoulou, Mourad and Matarić separates responsive conversation from slower deliberation. It also acknowledges coordination problems, including a conversational component working with stale state. It gives us a relevant design precedent, not a guarantee about the speed, cost or correctness of this service.⁴

For this visitor, the practical rule is simple. Reflect the correction now. Prepare the deeper work when useful. Offer it when it will help. If she is reading the morning page, leave it still; tell her another page is ready and let her open it. Waiting honestly is better than supplying an unearned conclusion to keep the conversation moving.

The conversation can be immediately useful without having to produce its deepest conclusions immediately.

The folder includes what is still owed

By this point there are source notes, an old role sketch, a corrected role, a possible morning and a current understanding. The first sketch stays, clearly marked superseded. Its reason for replacement is useful: it prevents a later helper from proposing staff management as though nobody had thought of it before.

Keep the reasons the earlier pages changed. A folder with only the latest summary loses the route by which the person arrived there. Equally, a pile of undated drafts can mislead its next reader. Current and superseded need to be obvious, with the sequence and reasons intact.

The next visit can therefore start with the actual unfinished work: the role interested her, the follow-through responsibility remained open, and she wanted material for a conversation with her partner. Ask whether those things have changed. Do not make her reconstruct the session to earn continuity.

Run it with paper

A friend can try the underlying method without any software. Put one appropriate piece of work between you. Let its owner explain a particular incident; the listener writes what they think happened and reads it back. Keep the evidence note separate from the interpretation.

On a fresh sheet, write a possible contribution or working day that follows from the incident. Ask the person to mark the part they would choose and the part they would refuse. Ask what makes the refused part wrong. Write the changed description on another sheet, keeping the earlier one and its reason for replacement.

Finish by agreeing what is established, what remains a guess and whether anyone owes a next piece of work. Give the sheets to the person whose experience they describe. A polished profile is unnecessary. An inspectable change is essential.

Our visitor has one more piece of work to request. She wants a page she can take into the conversation at home.

The page owed to our fictional manager cannot be prepared properly until we know which conversation she wants to have. A business partner needs to see how the work might be divided. A life partner needs to see what this possibility might ask of their shared life. Sometimes the same person needs both pages. They still answer different questions.

It is easy to treat “I need to discuss it with my partner” as a delay before a decision. I want us to treat it as a request for useful work. What does the partner need in order to understand the possibility, question it and reasonably decline?

Our job every time we interact is to be valuable enough to earn the next interaction.

That includes an interaction we will not attend. A page that makes the conversation clearer has done something useful. A page that leaves out the difficult parts so the visitor can win agreement has betrayed the work already done.

The business conversation

The first specimen continues the manager's corrected role. It is written for a possible collaborator to inspect, with no assumption that they will accept the unclaimed responsibilities.

Acquisition matters because a role built around booked conversations has no work to do until appropriate conversations exist. Commercial authority matters because explaining a service can slide into promising a delivery date or an exception. Follow-through matters because her preferred stopping point depends on another person accepting the handover.

Administration matters for the less glamorous reason that it still happens. If the pair never discuss it, it may land with whoever is most conscientious. A responsibility picture should expose that possibility before goodwill turns into a permanent job.

The page also protects the business partner from being cast as the answer to every blank. They may have useful knowledge without wanting to operate a service. Their refusal can reveal that the proposed shape does not fit either person. That is a successful conversation if it prevents an obligation neither intended to take on.

The conversation about everyday life

A life partner may have little interest in the responsibility chart. They may want to know whether this means more evenings at a screen, less certainty about income or another commitment in a week that is already full. Those are central questions, not resistance to an exciting opportunity.

Suppose the partner points out that the apparently free morning is when the visitor already handles a shared obligation. The table changes again. The next version cannot keep describing that morning as available. It should record that availability is unresolved and that the existing commitment remains in place.

The discovery partner does not need to adjudicate the household. They need to stop building suggestions on an assumption the household has rejected. Warmth here is taking ordinary life seriously.

The page should support an honest conversation, not become a personalised closing argument. One way to test it is to ask whether it contains enough information for the partner to disagree intelligently. If all the reasons to continue are vivid and all the costs remain vague, it needs another draft.

Invitation is specific

Mentioning a partner authorises no message. The visitor reads the proposed page, chooses what belongs in it and decides whether to share it. The service should make that choice concrete: this page, for this recipient, to support this conversation.

The same applies to an expert. Our manager might want someone experienced in service operations to examine the handover boundary. She could share the corrected role and responsibility picture with a question about where follow-through usually breaks. The expert can begin at the interesting uncertainty rather than asking her to perform her whole history again.

There is no need to include the page about home life, the entire source archive or every discarded role. Selecting a useful packet is part of the handover. Before sending, she should see exactly what it contains. Once someone receives a file, the service cannot promise to make them unsee or return every copy of it.

Luft's account treats disclosure and awareness as sensitive features of a relationship, rather than goods to maximise indiscriminately. That is consistent with giving the visitor control over which parts of the work enter a new conversation.²

The maker does not get a permanent seat

Songbird's ordinary staff should mostly be unable to browse these personal folders. Permission for the agents and processors required to deliver the service needs to be explained separately; it must not imply open access for anyone who works at the company.

That design choice has consequences. We cannot rely on a helpful founder noticing every poor unattended experience by reading everybody's pages. The service has to carry its care in the work itself. When we learn across folders, we should use consented, de-identified patterns that do not expose someone's story. A recognisable personal account does not become marketing material because its name was removed.

The partner pages now extend the usefulness of the folder beyond its original table. But they also carry the original interpretation into new hands. That makes the next question unavoidable: how do we know the description deserves to travel?

A partner can read a beautifully written page and still be reading a mistake. The folder needs a stronger test than whether its owner enjoys the description.

Bertram Forer's classroom demonstration gave students the same personality sketch while presenting it as individual feedback. They accepted it as personally accurate. The Barnum effect, also called the Forer effect, names this trap. The experiment is an enduring warning about personal validation: agreement with a description does not establish that the description distinguishes its subject. It was not an experiment on AI, but the warning applies directly to the temptation to write pleasing profiles.⁵

The response I would love someone to have is that they have never been described that way before, and the description is true. That is a target experience, not a customer testimonial. The word true is where the work sits.

Three moves, with different jobs

The inventory can contain titles, stories, work samples and preferences. The outside eye asks what those things reveal and where the revealed capability might be useful. The retelling joins that reasoning into language another reader can use, carrying the limits along with the insight.

Skipping the middle move produces attractive prose with no defensible comparison. Skipping the last leaves the visitor with an interesting conversation and the same writing chore they arrived with.

The nurse's evidence packet

Here is a second, fully fictional specimen. It describes no actual nurse, family or patient. Its purpose is to show what the method would have to do, including how its conclusions can fail.

The professional title provides context. It does not, on its own, establish any particular trait. The two incidents are more useful because they contain an observation and an action. The preference prevents us from turning competence into an unwanted future job.

The horoscope version

That paragraph feels supportive. It makes almost no use of the packet. Compassion may be present, but we have not shown it. Trust is attributed to people we have not heard from. Natural leadership ignores her stated aversion to staff management. Readiness for a bigger role is invented.

The same paragraph could be handed to a teacher, an administrator or a neighbour who helps people with forms. Substituting a nursing reference would not repair it. It needs claims that depend on this person's evidence.

A retelling has to be specific enough that it could be wrong.

A first account that can be inspected

This account makes several commitments. It says she initiated particular questions. It identifies a pattern across the two stories. It proposes a recipient for the capability. It preserves a preference. Every one of those statements can be challenged separately.

The possible recipient is grounded in a real problem shape. AHRQ describes care coordination in terms of deliberately organising care activities and exchanging information among the people involved. Bajorek and McElroy's primer on discharge planning describes the complexity of transitions and the work needed to make them safer. Together, they establish that coordination and understandable transitions are substantive concerns.⁶⁷

That leaves the outside eye with unfinished work. Compared with a generalist communicator, her experience might help identify questions the communicator would miss. Compared with another experienced nurse, it may be entirely ordinary. The folder should name the comparison it intends to make, instead of allowing the word valuable to conceal the missing work.

Even the tabs need correction

A folder's tabs can do a powerful piece of retelling. They make patterns visible at a glance. They can also exaggerate faster than a paragraph because they have so little room for qualification.

The first candidate implies measured speed and comprehensive understanding. Neither exists here; ten minutes must not become a claim about an adequate clinical assessment. The second makes a claim about other people's feelings. The third turns two incidents into knowledge of an entire system. Each revision preserves something useful while giving up the unsupported part.

The result is less grand and more informative. A future reader can ask about a daily routine or an unclear handover. It is much harder to ask a useful question of “practical wisdom”.

Try to break the account

Suppose she corrects the first inference. In the handover story, the ownership question was already on a checklist. She followed it carefully; she did not invent the question. That does not make the incident worthless. It changes what it shows. Consistent use of a method can matter, but it cannot be presented as evidence that she originated the method.

Now suppose the second story remains as described: she independently asked about the ordinary day and noticed the mismatch. We have one supported example of that initiative, not two. The account must become narrower.

This is moderation with consequences. It changes the substance, keeps the correction and leaves an honest next question. Merely replacing exceptional with strong would have left the original error intact.

Make the comparison small enough to answer

The next inquiry could examine a team that produces explanations for families dealing with several services. What does its current material already cover? Where do users still become confused? Could an experienced generalist identify those gaps just as well? What would this nurse notice that the team currently misses?

Only after inspecting that actual need could the outside eye say something useful about relative value. A willing recipient and a suitable task would still not establish a whole market. For now, the folder holds a proposed comparison and the evidence needed to settle it. No outreach or test is implied by writing the question.

Two readers, the same truth

The person needs an account that feels recognisable because it respects what happened. A later agent needs enough structure to avoid making the same mistakes again: source notes, the current version, the withdrawn inference, preferences, open questions and permissions.

If the agent receives only a polished paragraph saying she is a natural leader, it may keep proposing leadership roles. If it receives the corrected account, it can ask whether a particular task fits without converting interest into commitment or a current preference into a permanent identity.

The full description now has enough substance to carry. The next temptation is to compress it into a name.

Compress the nurse's account far enough and you might arrive back at nurse. That would not be a failure. The title could remain familiar while the description underneath it becomes far more useful. The purpose of the folder is not to give everyone an impressive new identity.

Sometimes, though, a different name changes what a person is willing to recognise in themselves. I know because it happened to me.

It took me a long time to say scientist

I was doing work that was not quite writing AI code. There was thinking about how the business should work, developing the doctrine, working through methods and trying to make the ideas coherent enough to use. I was doing it, yet I could not see clearly enough how valuable that work was.

My co-founder treated me as the IT person. That description brought the mismatch to the surface. I did not want the title. Rejecting it forced me to work out what I did want to be called, and what work I wanted that name to stand for.

It took me a long time to put scientist beside my name. Chief scientist was the label I eventually wanted. When I found it, something changed in how I thought about my contribution. I began leaning harder into the doctrine, the thinking and the ideas. I could see that work as my part, rather than as an awkward collection of things around writing code.

The discomfort matters to the story. I had not produced a neat account of my value and then selected a title from it. Someone else's inadequate account gave me something I could refuse. The refusal helped me see the difference.

That experience is evidence about my own recognition. It does not mean a title confers a qualification or proves the quality of someone's work. It means the words I used had been too small for the work I wanted to own. A better description gave me a clearer direction.

A label is a retelling in one word.

Offer a name with its working day attached

The useful part of a candidate label is the activity it brings to mind. Without that, choosing between labels can become an exercise in status or taste. Our fictional manager provides a practical way to test them.

Take salesperson. She might reject it because it suggests persuading people who have not asked for help. Another person might hear attentive commercial advice in the same word. The partner needs her meaning, not a dictionary defence of the title.

Customer guide may feel closer until she asks whether she will receive every call from that customer afterwards. Service interpreter may capture the explanation but sound detached from getting the next action accepted. The best answer could be to keep the sentence they already wrote and defer the name.

Ask which activity makes the label wrong. That question lets someone distinguish a word they dislike from a responsibility they refuse. If the concern is ongoing availability, revise the boundary. If the concern is persuasion, preserve the difference between explanation and pressure. The correction should improve the account even if no label survives.

Visible candidates give memory and judgement something to engage with. Budiu's discussion of recognition and recall supports the value of providing cues rather than demanding unaided retrieval. It does not tell us which name is correct, and a strong reaction is still something to understand rather than score.³

Do not manufacture the wrong answer

My co-founder's description helped me by being wrong. That is not a reason to provoke visitors with deliberately belittling labels. The service can offer sincere possibilities with their limits visible. It should never engineer discomfort so that a later, flattering answer feels like rescue.

The visitor can reject all the candidates, leave the question open or decide the naming exercise is not useful. Nothing important has been lost if the folder already contains a precise account of the contribution. The partner has still done the writing.

A useful final sentence for our manager might be: “I help customers understand suitable options and leave a clear handover, with responsibility for delivery accepted by someone else.” That sentence has more practical content than a title alone. It tells a collaborator what to ask next.

The name remains revisable

A later role sketch may reveal a new limit. Suppose a customer-explanation role includes negotiating exceptions. The manager could like the general shape while refusing that authority. Her folder would then need a more precise boundary around what she can explain and what she must refer onwards.

Keep the earlier version with the reason for change. Mark the new one as current and record when the preference was expressed. A later reader can see that the account developed through experience rather than treating every revision as inconsistency.

Declining that role does not erase her past capability. Nor does accepting a label require her to accept every job associated with it. The folder should help her exercise judgement about possibilities, including the judgement to leave them alone.

Once the words are useful, they need to travel. A name held inside somebody else's service is still dependent on that service. The next act is to put the account into the person's hands.

Imagine the person arriving and the person leaving. Keep the posture the same. Keep the expression the same. Resist the temptation to make the first figure small and defeated, the second suddenly confident and successful. The only thing we need to add is the packed folder under their arm.

My image of the folder comes partly from accounting work before software became the centre of everything. A physical client file brought the particular client's circumstances and unfinished work to the next person handling it. You could inspect what had happened and what still needed doing. Songbird inherits that sense of an entrusted, continuing piece of work. Here, the decisive change is who takes it away: the person it describes.

I have the archive. I still need the account.

I have 28 years of my own email and documents, having worked for myself for most of that time. There is a great deal of evidence of what I have done. I have only marginally turned it into a useful account of what I know.

That is an awkward admission from someone who keeps telling other people to write down their intellectual property. I have repeated the point to people around me. It has not caused the folders to appear. Including me, there are three people close to this conversation who understood the argument and still had not done the work properly. That is an observation about us, not a survey of a market.

During the last year, writing ebooks and putting the material into a form I could use has helped. It gives my thinking a surface. I can return to a developed explanation instead of reconstructing it from fragments of conversation. But a year of writing does not suddenly organise decades of emails, decisions and work.

An archive offers abundance without selection. I can search for a project name and find documents. That does not tell me which recurring request reveals a capability, which incident deserves a page or which old interpretation I no longer believe. More material can make the selection problem larger.

The writing is the crucial part.

Knowledge exists before it is written. People exercise it every day. Writing changes who can inspect it, how easily it can be reconsidered and whether another person or agent can use it without asking its holder to start again. That is the part I keep advocating and only partly doing.

So I do not want to send visitors home with instructions to produce their own folder. We have already seen how much interpretive work sits between a story and a usable account. Doing that work with them is the gift.

What would actually be handed over?

Take the manager's specimen. The opening page should say what the folder is, which account is current and where to start. It should not require a tour from the person who made it.

The complete folder and a sharing packet are different objects. The complete folder is for its owner. A packet contains the pages chosen for a particular conversation. The manager might give a collaborator the role and responsibility pages while keeping her household discussion private. Export must not silently treat every page as equally shareable.

A file, in the ordinary sense

The delivery I would ask us to build includes a self-contained HTML copy that opens in a browser, plus an editable Markdown or plain-text copy. Images and other essential material belong with the delivery, rather than depending on a private link that later expires. A printable copy can be useful for a real kitchen table; it should not be the only editable record.

Markdown is useful here because the text remains human-readable and editable. The informational RFC describing its media type also recognises variants. That is a reason to keep the content simple and check it in the receiving tools, not claim that a format name guarantees perfect interoperability.⁸

Consider the owner's ordinary next action. She opens the text copy, changes her availability and saves a new version. She can give that account to another adviser or use it with a tool she chooses. The important distinction between evidence, interpretation and open questions should remain legible after the decorative presentation has gone.

Before calling the handover complete, check that the current account can be read offline, that text can be copied and edited, that attached material is present and that the selected sharing packet contains only the intended pages. Check that superseded claims remain marked as superseded. These are practical tests of the promise.

Ownership must survive leaving

The files alone do not answer every question. Before offering the service, we need plain terms saying what Songbird retains, for how long, which authorised systems or people can access it and what choices the owner has about those retained copies. Those terms must match the system we actually operate. A proposed folder design is not an implemented retention policy.

But the immediate ownership test is concrete. The visitor can keep, edit and use the delivered account without buying another interaction. They can choose whom to show it to. They can stop working with us while the useful thing remains in their possession.

That freedom answers the unease behind handing an idea or a personal history to a platform. The service demonstrates its intention through a completed handover. It does not ask the visitor to infer generosity from a promise about some future stage.

Now the folder is under the person's arm. Making it, checking it and handing it over all cost something. If the first gift is free, that cost belongs on our side of the table.

The delivered folder has already consumed resources. Someone prepared the material, paid for the computation, checked the account, handled corrections and completed the handover. Calling it free changes who pays. It does not remove the bill.

I want the first part to be free. That still leaves us responsible for its cost. The visitor should be able to understand both halves of that statement: the gift is complete, and the company has a reason to fund it.

Three honest exits

One person may want to pilot a company. The folder helps them enter that later conversation with a better account of what they can contribute. Nothing in the first gift should pretend that decision has already been made.

Another may want to lend know-how to someone else's work. They may have no wish to operate a company. Their experience can still matter, subject to a specific invitation and their choice to contribute.

A third may take the folder and leave. They may use it to explain themselves in a job conversation, discuss a change at home or simply keep it. They owe us no further participation for the work to count as delivered.

These are three honest exits only if the experience treats them as legitimate. A membership invitation inserted before the folder is complete changes the conversation. A download that appears only after a sales call changes it again. The commercial offer belongs after the gift and requires a fresh choice.

The new freemium

Conventional freemium often withholds features until someone pays. The new freemium here funds an early piece of useful work because delivering it can improve the company. There are three possible returns: sensor, supply and marketing.

Sensor. With appropriate consent, de-identified patterns can reveal what people struggle to recognise or picture. If several visitors cannot evaluate a role until they see an ordinary working day, we can improve how we prepare those pages. The useful learning is the pattern and the change it suggests. We do not need to appropriate a recognisable person's story to learn it.

Supply. Some people will choose to continue as founders or contributors. Their folders can make a later, permissioned conversation more grounded. Receiving a free folder does not enrol someone in a pool that others may search or contact. Their personal knowledge is not an unspoken payment.

Marketing. A useful gift may earn a recommendation. A visitor might show a selected page to a friend and explain why it helped. That possibility is more credible than a claim about being human-focused, but it remains a possibility to measure. We cannot put referrals into the budget as though gratitude guaranteed them.

Each return can fail. Patterns may be too thin to use. Few people may want to continue. The work may be appreciated without being recommended. Those failures would not make the recipient indebted; they would make our funding hypothesis weaker.

Scale the generosity, learn the craft

A founder who gives every lead an hour already demonstrates a willingness to do useful work before payment. That habit is worth preserving. It does not establish that the personal folder has been mastered or that automated delivery will be cheap.

Paul Graham's account of doing things that do not scale argues for direct, manual work with early users as a way to learn what a company should make. It is relevant to the first folders because the learning comes from the actual service. It is entrepreneurial advice, not a measured cost model for this one.⁹

We should therefore budget for the full work while learning which parts become easier. Cheap generation can coexist with expensive checking. A fast first draft may create more correction work than a slower, better-prepared one. The cost of a useful completed folder is the number that matters.

There is no measured production cost or justified numerical weekly cap available for this proposal. Writing “a few dollars” would conceal the largest unanswered questions. The responsible next step is to measure the first hand-made folders and reserve enough to finish every accepted one.

A cap that protects the promise

Before accepting a week's new folders, subtract the money and review time already committed to unfinished work. Divide the remaining funded allowance by a conservative allowance for completing a new folder. Separately, calculate how many new folders the available review capacity could support. Admit no more than the smaller number.

A visitor need only hear that places are limited because the work receives proper attention. Internally, the limit needs real inputs and an accountable owner. An aspiration to serve everyone is no reason to accept work the company cannot finish.

What the review has to decide

Look at the spread of completed-folder costs, including difficult cases, rather than celebrating a cheap average. Look at whether owners could correct the account and actually take it away. Count the three exits without treating departure as a defective conversion.

Then examine the proposed company returns. Did a consented pattern lead to a specific improvement? Did anyone voluntarily choose a further contribution? Did recommendations bring people who wanted this service? Pages generated and conversations held are activity measures; neither establishes that the subsidy is buying useful learning.

At the review date, the service owner must be able to reduce or stop new admissions if quality falls, completion costs exceed the funded reserve or repeated work produces no useful learning or other intended return. Already accepted folders remain obligations to complete. An export paywall is not a remedy for an incorrect budget.

Three ways to betray it

The first betrayal is flattery. It can make approval cheap and leave the person with a description that fails the moment someone asks for an example. The could-be-wrong test protects the gift from that temptation.

The second is a reluctant goodbye. A folder that becomes difficult to obtain when the owner declines a next step reveals the real bargain too late. Departure should receive the same competent completion as continuation.

The third is an account that feels machine-made because it has not attended to the person: generic language, ignored corrections, a shiny title unsupported by the material. Lowering inference cost does not justify handing over that product.

I would be proud to build a business that did this for someone and let them leave. That feeling is why I want the service to exist. The budget, evidence and stopping rule are how we take responsibility for making it exist.

Even a sound budget cannot tell us how to make the first truthful folder. That craft has to be learnt at an actual table.

A funded place is permission to begin the work. It is not evidence that we know how to finish it well. The first folders need a human maker close enough to notice when a pleasant conversation is producing a poor account.

There are two crafts in our starting material. A founder already helps people clarify an idea by listening, finding its useful centre and retelling it coherently. Personal retelling asks a different question of the same conversation. What does the idea, and the experience behind it, reveal about its holder?

Skill in the first gives us a useful beginning. It does not let us skip making examples of the second. Graham's case for manual early service matters here because direct work can reveal what needs to be built before a company automates its assumptions.⁹

Three customer-zero paths

Begin with a founder who works by talking. Ask for an incident rather than a self-description: a time someone brought a confused idea, what the founder noticed and how the conversation changed it. Write the account while the founder corrects it. The finished object should identify a contribution and its limits without requiring the founder to do the writing afterwards.

The danger is familiarity. A colleague can silently fill gaps from years of knowing the person. Mark any fact that came from prior knowledge rather than the session. The learning question is whether the method elicited the material or the relationship supplied it invisibly.

Then use my documents-first path. Start with a selected, appropriate part of my own archive, not a wholesale upload. Look for recurring requests: what did people repeatedly ask me to help with? Choose a small set of examples and ask me to explain what the documents leave out. A request shows that someone sought help; it does not automatically prove I gave good help.

The output should connect an account of my work to particular evidence, with corrections I have made visible. The failure to watch for is volume masquerading as understanding. A long summary of an archive can still miss the judgement that mattered. We should learn which documents earn a place on the table and which merely increase the reading burden.

After the founders, work with a friendly outsider. This person should not need an archive or an existing relationship with the writer. Invite one story they are comfortable discussing. Build the first page from that material alone. Ask them to correct it, then give them the result.

The cold start tests whether our language makes sense without the company's private vocabulary. If the outsider cannot tell what the page claims, or politely agrees with a generic account, the method has more work to do. Their willingness to participate is not a testimonial, and their approval needs the specificity test already established.

A care floor, before the session

I want a clinically trained adviser involved in setting the care floor. That is a proposed role, not an appointment or an approval already obtained. The following is a draft standard to take into that work, alongside the actual hand-made examples.

Luft's warning against forced awareness is especially relevant to the second point. More disclosure is not automatically progress. A respectful boundary may be one of the clearest things learnt in a session.²

Care should not make the table feel like an examination. A visitor can be welcomed, surprised by a well-chosen page and allowed to enjoy the work. Generosity is compatible with clear limits. What it cannot do is bypass them because a more intimate conversation would make a more dramatic story.

The adviser should be able to inspect how the session actually responds to refusal and uncertainty. A polished policy alone cannot show that. A real corrected page, viewed with the participant's permission, can.

Give the AI an apprenticeship

First, make complete folders manually. Keep the evidence, the proposed account, the person's corrections and the final delivery together, with permission for any learning use. The examples need to include the awkward parts: a rejected inference, an unhelpful question, a page that had to be rewritten.

Next, let AI prepare drafts while an authorised person checks the work. Compare the draft with the actual evidence and observe whether corrections survive into subsequent pages. A reviewer should not have to guess which claims were invented to make the account read smoothly.

Only then consider a limited unattended experience. It must demonstrate that it can produce specific, bounded accounts; accept corrections; preserve sharing choices; complete an accessible export; and stay within the funded completion reserve. Human review of personal material needs the person's permission. Secret spot-checking is incompatible with the access promise.

A smooth conversation is insufficient evidence for that transition. Forer's demonstration reminds us why satisfaction with personalised-sounding prose cannot carry the whole quality judgement.⁵

A fifteen-minute experiment to run next

The immediate experiment is smaller than a full folder. Take a willing person who has brought an idea and spend fifteen minutes understanding a story behind it before returning to the idea. This is a proposed exercise; it has not been run for this book.

Before the timed conversation, explain its purpose, agree the topic and decide whether notes or a recording are appropriate. If recording, obtain both participants' consent and agree who can use it. Write a brief starting account of the idea so there is something concrete to compare afterwards.

For the opening few minutes, ask what the person saw that made them want to fix this problem. Find one incident. Who was trying to do what? Where did it go wrong? What did the person notice, and what did they do? Avoid letting the conversation dissolve into a general history of the industry.

Use the middle of the session to distinguish observation from interpretation. Ask what a less experienced person might have missed, then ask for the incident that supports that difference. If the person cannot supply it, leave the comparative claim open. Ask which part of the work they would choose again.

Before the end, write one specific account of the contribution you think the story shows. Read it back. Ask what is wrong or missing, change it and let the person keep the corrected version. Return to the starting description of the idea and record whether you now understand it differently, and exactly which evidence caused the change.

The output is modest: a starting idea description, an incident note, a corrected personal observation and a changed or unchanged understanding of the idea. One conversation cannot establish that story-first interviewing causes better outcomes. It can expose a useful question, an unsupported inference or a failure of the proposed method.

The two-minute slice

A homepage could offer an even smaller encounter: one chosen story, one provisional observation, an invitation to correct it and a copy to keep. The promise must stay that small. Two minutes should not pretend to deliver the whole folder or discover a person's defining capability.

The useful test is whether the observation depends on the story. If it could be sent unchanged to the next visitor, revise it. If it assigns a grand trait from one incident, narrow it. If the visitor corrects it, the copy they receive must contain that correction.

No membership request is needed inside this slice. It has a job already: let someone experience a little of the work and decide whether more would be useful.

These early pages will also do something for their makers. They will put the proposed service where a founder, adviser or craftsperson can point to it, reject a part and show us what needs to change.

Put the first proposed folder in front of the people who will build the service. The discussion can become unusually precise. Someone points to an unsupported claim. Someone notices that a correction vanished from the next page. Someone asks why the business partner has been assigned all the work the visitor declined.

Those responses are more useful than another general conversation about making the experience excellent. They show where an intention has failed to reach the work.

The same move appears beyond personal discovery: give someone a concrete proposition, let them react, then preserve what their reaction teaches. Polanyi's account of tacit knowing explains why being able to do something need not include a ready verbal account of how it is done. It does not promise that every part of expertise can be extracted into sentences.¹

An expert finds the missing condition

Imagine, as an illustration, an experienced service operator reviewing a deliberately incomplete alternative to the manager's handover page. This version says her part ends when she sends the note. The operator rejects it: sending does not establish that anybody has accepted responsibility.

The useful response is to ask for the condition that would make the handover complete, and an example of what happens without it. The operator might describe a shared inbox everyone can see but nobody owns. Now there is something to write: the receiving person accepts the next action, or the handover remains open.

The correction has produced a more inspectable rule. It still needs testing against other situations. Some workflows may have a different, explicit acceptance mechanism. One expert's confident refusal is a source of a proposed distinction, not proof that the distinction is universal.

A craftsperson chooses between two pages

Now give a craftsperson two versions of the nurse's account. One opens with a flattering title and places the source note far below. The other puts the specific observation beside its evidence and makes the correction easy to see. Ask which page they would hand to the person, and why.

Suppose they choose the second because its reader can check the claim without hunting. That reason is more useful than “make it cleaner”. It suggests a principle to try on another page: keep the evidence close enough to the interpretation that the reader can inspect the connection.

Ask about the cost as well. A page can become so crowded with qualifications that the person cannot find the sentence meant for them. The craftsperson may choose a clear main account with the evidence immediately available, rather than every detail at equal weight. The next example tests whether we understood that judgement.

Recognition cues help explain why a comparison can be easier to discuss than an empty request for criteria. Budiu's treatment of recognition and recall supports providing such cues. Choosing a page still expresses a judgement in a context; it does not establish a permanent design law.³

A co-founder can inspect the promise

A co-founder may struggle to engage with an unfinished explanation of how personal discovery will work. A proposed morning gives them something firmer. Here is the visitor. Here is the piece of work she brings. Here is the page we make. Here is the responsibility we accept when she corrects it.

They can now say that the preparation is too thin, that the next obligation has no owner or that the supposed free gift stops before the useful part. Those objections change the operating promise. They need not be interpreted as a lack of imagination or an inability to understand the theory.

In an illustrative review, the co-founder might point to the partner page and ask who finishes it after the visitor decides not to continue. Writing a named completion responsibility into the proposal turns generosity into a piece of work someone owns. The disagreement has earned its place in the company's understanding.

The company needs a folder

I have been writing doctrine because the company has the same practical problem I described in my archive. Much of the thinking exists in conversation, in fragments and in versions that have already moved on. The company does not yet have a complete folder of its own. Writing the doctrine is how it begins to get one.

That folder would hold the current decisions, their reasons, the examples that support them and the questions still open. It would say who owns the next piece of work. When a decision changes, the earlier version and the reason for changing it remain available.

This does not require everybody to agree with a final account forever. It makes disagreement productive. A co-founder can point to the actual promise, offer contrary evidence and help revise it. The next conversation begins with the changed page instead of somebody's recollection of what was said last week.

The analogy has a boundary. A company's shared doctrine is not permission to share a visitor's private folder. Each object has its own owner, readers and purpose. The common method is visible, correctable understanding; the access rights do not transfer with the metaphor.

Where this sits in the larger work

I think of this as one application of the outside eye. The wider question of eliciting judgement at different scales deserves its own treatment. Here, the method earns its place by serving a particular person and finishing a particular gift.

AI as the Interface supplies the idea of changing useful material together. The Fast–Slow Split and Personal Agents' Three Jobs inform the separation between a live exchange and deeper preparation. The Proposal Compiler and Marketplace of One sit nearby in the larger work of making something concrete enough for someone to recognise. Their machinery is not needed to begin at this table.

Begin with an incident the person is willing to discuss. Put a sentence about it where both of you can see it. Keep the source and the interpretation distinct.

Perhaps the sentence says they solved the problem by taking charge. They cross out taking charge. They explain that they asked the question nobody had asked, then found the person who could act.

You change the sentence. The new page describes that contribution and keeps the limit. They read it again, make one more adjustment and ask for a copy.

Give them the page.