Idea Board

Who it's for

Librarians, community-centre staff and volunteers who are asked AI questions but have no training to answer them.

How it works

A printed facilitator guide, five practice cases with planted errors, a one-page checking routine and a sign-up sheet. New cases are emailed monthly.

The AI angle

An assistant generates fresh practice answers with deliberate, realistic mistakes for volunteers to catch. A human reviews every case before it's sent.

First step

Pilot the kit at three branches and measure how many attendees catch the planted error by week four.

Seed: libraries, retirement, fact-checking

Who it's for

Owners of cafés, bakeries and salons with 3 to 15 staff who currently plan rotas with a pencil on Sunday night.

How it works

Staff enter availability and 'never' rules on their phones. The owner gets a draft rota plus a checklist showing pass or fail for every rule, and hours are totalled by a formula, not the model.

The AI angle

The model solves the puzzle of who goes where; a second, fresh check hunts for broken rules. The arithmetic stays in plain spreadsheet logic.

First step

Run the NEVER-list prompt by hand for five local shops for a month and count how many rule breaks the check catches.

Seed: baking, small business, scheduling

Who it's for

Students and junior developers whose projects were built with coding assistants.

How it works

Connect a repository. It interviews you about decisions, finds places where you changed or rejected AI suggestions, and drafts a 'decisions and mistakes' section for each README.

The AI angle

The assistant asks questions and drafts; the developer's own explanations are the substance. It never invents stories about the code.

First step

Run five students through it by hand, with me asking the questions, and show the results to three hiring managers.

Seed: recruiting, junior developers, careers

Who it's for

Science teachers and school IT leads choosing which AI tutor to allow in class.

How it works

Teachers contribute real student questions with model answers and a four-part rubric. Any tutor can be run through the set blind, and results are published as plain-language scorecards.

The AI angle

The benchmark itself is human-written; AI only runs the candidate tutors. The rubric rewards asking 'what have you tried?' over solving the problem.

First step

Publish my Year 11 class's 30 questions and rubric, and invite five other teachers to add ten questions each.

Seed: chemistry teaching, evaluation, students

Who it's for

Hobby woodworkers who watch build videos and then spend an hour pausing to write down measurements.

How it works

Paste a video link or upload the transcript. It pulls out every dimension, material and tool mentioned, builds a cut list and links each line to the moment in the video.

The AI angle

Transcript extraction with timestamps. Anything ambiguous, like 'about this long', is flagged rather than guessed.

First step

Do it by hand for ten popular build videos and ask viewers in the comments whether they'd use the result.

Seed: woodworking, YouTube, DIY

Who it's for

Small freight forwarders and importers who lose days to rejected paperwork.

How it works

Drag in the documents for one shipment. The tool extracts weights, values, codes and names, lines them up in a table and flags every mismatch with the page it came from.

The AI angle

Document extraction with citations, so every flag is traceable. It never edits documents, it only points.

First step

Collect 50 past rejected shipments and check whether a simple extraction table would have caught the reason for rejection.

Seed: logistics, spreadsheets, West Africa trade

Who it's for

Parents and hosts cooking for someone with an allergy who don't want to give up favourite recipes.

How it works

Paste a recipe and choose the allergens to remove. You get a remixed version with every substitution highlighted, the reason it works and a warning wherever a swap changes texture or rising.

The AI angle

The model suggests substitutions; a curated, human-reviewed table of safe swaps constrains it. It reminds you to check labels every time.

First step

Test fifty substitutions with three home bakers and a dietitian before building anything.

Seed: cooking, allergies, family

Who it's for

Food banks, shelters and community groups that run on volunteers who don't want another app.

How it works

Volunteers get reminder texts the evening before. Replying SWAP triggers messages to suitable people on the waiting list; the coordinator approves the match with one tap.

The AI angle

The assistant matches skills and availability and writes friendly messages. A human approves every swap.

First step

Run it manually for six weeks, with the coordinator doing the matching from a spreadsheet, to see whether no-shows actually fall.

Seed: food bank, volunteering, SMS

Who it's for

Managers and finance teams surprised by AI invoices they can't interpret.

How it works

Connect usage logs. It groups costs by feature, shows the average request broken into question, attached context and answer, and suggests the biggest saving.

The AI angle

An assistant writes the plain-language summary, but the numbers come straight from logs, never from the model.

First step

Produce the report by hand for my own team's last three months and present it to my boss as a test.

Seed: fintech, product management, AI costs

Who it's for

Doctoral students in the final year, especially those without much supervisor time.

How it works

Upload a chapter and your field. It extracts every claim with a page number, finds claims with thin support and generates examiner-style questions ranked by how likely they are to come up.

The AI angle

Extract-then-question, so every question is anchored to a quoted passage rather than a vague impression of the chapter.

First step

Try it on three finished theses with their real examiners' reports and see how many of the actual questions it predicts.

Seed: PhD life, academic writing, biology

Who it's for

Narrative designers and small game studios with branching dialogue and no QA budget.

How it works

Export your dialogue tree. The agent plays through each path, logs what the player knows at each node and flags paths where a needed clue never appears.

The AI angle

A small local model handles many cheap, repetitive checks so unreleased scripts never leave your machine.

First step

Build a script that walks my own detective game's tree and see whether it finds the dead end I already know about.

Seed: indie games, narrative design, local models

Who it's for

Owners of salons, food trucks, contracting firms and shops who want practical help, not theory.

How it works

Each session opens with a five-minute skill (checking answers, privacy, prompts for quotes or reviews), then an hour of working on the task each person brought, with a facilitator circulating.

The AI angle

Assistants are the tool on every table; the course teaches habits that survive tool changes.

First step

Run the first cohort at my community college in January and track how many tasks are finished in class.

Seed: teaching, small business, community college

Who it's for

Product teams shipping AI features who only have analytics, not insight into trust.

How it works

Participants log a 30-second voice note whenever they use the feature: did they trust it, check it or ignore it, and why. Researchers get a weekly synthesis with quotes.

The AI angle

AI transcribes and clusters the notes; the research team writes the insights. Every cluster links back to the original recordings.

First step

Run a two-week diary study with eight people on one feature and compare the findings with what the analytics said.

Seed: UX research, trust, AI features

Who it's for

Organisers of monthly repair events where queues build up and fixers run out of time.

How it works

People book a slot by uploading a photo and a short description of the broken item. The tool suggests a category and time estimate, and the organiser builds the schedule.

The AI angle

Image and text classification gives a rough estimate; volunteers correct it, and their corrections make it better over time.

First step

Collect a season's worth of repair logs from one café to see whether item type actually predicts repair time.

Seed: repair cafés, sustainability, volunteering