Chapter 07

Things I Haven’t
Done Yet

A few useful things. A few improbable ones.

Proposals for unwritten pages. Nothing built or promised here yet; the priorities are starting points to test.

Worth building, if the test holds.

Start with one ringing phone.

The agent that picks up

Test first · Unbuilt concept

An AI agent that answers calls, handles routine questions, books a visit, and passes the conversation to a person when it gets complicated. Later, it could make requested callbacks and appointment reminders.

Listen
Confirm
Book or hand over

A narrow, measurable job: turn an unanswered enquiry into a correctly captured request or booking. A useful pilot could earn its place without becoming a general-purpose assistant.

Unfold the first pilot

Who it could help: A repair shop or home-service business that cannot always answer the phone.
Why give it time?: A narrow, measurable job: turn an unanswered enquiry into a correctly captured request or booking. A useful pilot could earn its place without becoming a general-purpose assistant.
A small first version: One business, one phone number, one language and one calendar. Answer inbound calls using approved information, confirm details aloud, and send a short handoff note. Identify it as an AI assistant; keep a human fallback. Try outbound calls only with people who requested them.
What would make it worth continuing?: Run 30 supervised test calls, including interruptions, unclear dates and requests for a person. Then ask a pilot business to compare correct bookings, missed requests, correction time and cost per resolved call. Continue only if it is useful enough to pay for.
The difficult part: Medium–high. The difficult part is reliable conversation and booking, not making a voice sound convincing.

Feasibility reference: Twilio ConversationRelay →

Voice notes that do the work

Smallest useful pilotTurn a rambling voice note into a job request, a draft quote and the next follow-up—with the original words still attached.Unfold the ideaFold the idea

Who it could help: A small service team taking requests through voice messages.
Why give it time?: The value could be fewer details lost between a conversation and the job sheet. One repeatable workflow makes the benefit easier to test than an assistant that promises to do everything.
A small first version: Upload a voice note; extract the customer, task and requested date into an editable job sheet. Highlight uncertainty. Let a person approve every quote and outgoing message.
What would make it worth continuing?: Try 20 real, permissioned notes with one team. Compare manual entry time against correction time. Keep going if the team chooses it again the following week.
The difficult part: Medium. Noisy recordings, mixed languages and ambiguous names need careful review.

The missing piece

Validate the workflow firstA document desk that spots a missing attachment, an inconsistent name or an unanswered field before a submission leaves the office.Unfold the ideaFold the idea

Who it could help: A team repeatedly preparing the same kind of application or client file.
Why give it time?: A fixed checklist gives this idea a concrete job and a clear boundary. It is worth exploring if avoidable corrections are a recurring burden for a real team.
A small first version: One document type, a checklist supplied by its owner, and a review screen that points to the exact page behind each finding. The reviewer makes the final decision.
What would make it worth continuing?: Use 20 anonymised example files with known omissions. Measure missed problems and false alarms. Stop if reviewing the suggestions takes longer than checking the files by hand.
The difficult part: Medium–high. Scans, changing forms and sensitive information complicate an apparently simple checklist.

A sensible order: speak to a potential user, try the smallest version, then decide whether it deserves another week.

The stranger possibilities.

Some ideas earn their time through curiosity, craft or the people they bring together.

A pocket time machine

Worth a weekendAsk a private collection of notes and photos, “What was changing around this time last year?” Get a trail of original fragments, not an invented autobiography.Unfold the ideaFold the idea

Who it could help: Someone with years of scattered notes and photographs.
Why give it time?: An experiment in memory as something to wander through. Its value would be rediscovering an overlooked connection, rather than producing another summary.
A small first version: A local collection of 50 deliberately selected files, a timeline and search with exact source links. No background recording.
What would make it worth continuing?: Does it reveal one meaningful connection that ordinary folders would have buried? Can every answer be traced back to a real fragment?
The difficult part: Medium. Source fidelity and privacy matter more than the size of the archive.

The story that steps outside

Make one chapterA short story that unfolds during a walk. A crossroads becomes a choice; a familiar street becomes part of the fiction.Unfold the ideaFold the idea

Who it could help: Readers who would like to experience a story away from a screen.
Why give it time?: A chance to bring writing, place, sound and interaction into the same work. The story itself would have to earn the technology.
A small first version: One original ten-minute story with three stops and optional audio. Use manual check-ins, with a seated version that follows the same narrative.
What would make it worth continuing?: Ask five readers where they felt present and where the phone interrupted the experience. Expand only if they want another chapter.
The difficult part: Low–medium technically; strong writing and accessible routes are the real work.

A language you can keep

Explore with a communityA living collection of regional words: pronunciation, the phrase around them, and the story that makes a dictionary definition feel too small.Unfold the ideaFold the idea

Who it could help: Families and speakers who want to preserve words across generations.
Why give it time?: A small cultural project could matter without needing to become a large company. Contributors should decide what belongs and how it is shared.
A small first version: Twenty voluntarily contributed Sindhi words, original speaker recordings, reviewed translations and clear permission for each entry. No cloned voices.
What would make it worth continuing?: Invite speakers to correct the entries and add words of their own. Continue if they feel represented and want to help shape the collection.
The difficult part: Medium. Trust, credit and linguistic nuance are more demanding than the website.

Chess against your habits

Prototype the strange partAn opponent built around the mistakes that recur in your games, followed by a small practice set about the patterns you keep overlooking.Unfold the ideaFold the idea

Who it could help: Casual chess players with a collection of their own games.
Why give it time?: A more personal question than “What is the best move?”: what does this player repeatedly fail to notice? The experiment would be whether that makes practice stick.
A small first version: Import ten PGNs, use engine analysis to identify one recurring tactical pattern, and build five legal practice positions. Only then try an opponent that imitates those habits.
What would make it worth continuing?: Can the player recognise that pattern in an unfamiliar position a week later? If not, a convincing imitation alone is not enough.
The difficult part: High. A small game sample can suggest patterns that are not really there.

A room for unlikely collaborators

Try it without softwarePair two people from different crafts around one small question. A poet and a cartographer, perhaps. One week, one shared object.Unfold the ideaFold the idea

Who it could help: Curious makers who want a reason to work outside their usual discipline.
Why give it time?: The interesting part is what people make together. A manual experiment can reveal whether a platform would add anything at all.
A small first version: Invite six willing participants, agree on three pairings, and publish the resulting pieces only with their permission. Use a simple page and email before building matching software.
What would make it worth continuing?: Do the pairs finish something, and do any ask to collaborate again? Build a tool only after the exchange works on its own.
The difficult part: Low technical effort; high facilitation effort. It may be better as a recurring creative gathering.