← AI Prep

Change Log

What's changed in AI Prep, in plain language, newest first.

Changelog

This page lists what's changed in AI Prep, in plain language, newest first — for anyone using the tool, not a developer log.

[2.1.15] - 2026-09-18

Fixed

  • The Final System Blueprint could previously be generated even though a section flagged as possibly affected by a later change had never actually been reopened and checked — it just hadn't been looked at yet, and Real Builder was treating "not yet checked" the same as "confirmed fine." The Blueprint now correctly waits for that section to be reopened (and its check to finish) before it will generate.

[2.1.14] - 2026-09-18

Changed

  • Lateral Thinking now runs last, right before the Final System Blueprint, instead of right after the Capability Map. It now works from the complete, finished product definition and pushes back on the whole thing — missed opportunities, assumptions worth questioning, steps that could disappear or become AI-native — rather than just brainstorming extra features early on.
  • The sidebar no longer shows a vague warning triangle when an earlier change "might" affect a later section. Real Builder now identifies the exact answer(s) that actually need another look, tells you specifically why, and shows a precise count (for example "Workflows ↻ 1") — sections that aren't actually affected show nothing at all.
  • Every flagged answer can be resolved by keeping it as-is, updating it, using a fresh AI suggestion made specifically for that review, or marking it not relevant.
  • The Final System Blueprint now waits for every flagged answer (and any not-yet-reviewed Lateral Thinking capability) to be resolved before it will generate, so it's never quietly built from things you haven't actually looked at yet.

[2.1.13] - 2026-09-18

Added

  • Real Builder now includes four new stages after the Capability Map: Lateral Thinking (AI deliberately challenges the conventional product framing, with an adjustable "how far outside the box" level from 1-5), Capability Deepening (works out the detailed vocabulary complex capabilities need before they're implementation-ready), Financial Enrichment, and Reporting Enrichment (proposes the reports and dashboards the product probably needs). Lateral Thinking and Financial Enrichment can each be turned off per project.
  • A new Project Knowledge artifact captures everything Real Builder has learned about the product across every stage, research, and the final critique — and the Build Prompt is now written from that instead of directly from your raw answers, making it more coherent.
  • Ideas you accept from Lateral Thinking can automatically suggest new or changed capabilities on the Capability Map — but those proposed changes always show up for your review, exactly like any other capability decision, before they count as accepted anywhere else in the wizard.

Fixed

  • Reviewing (accepting or rejecting) a capability that came from an accepted Lateral Thinking idea now correctly flags only the later stages that actually depend on it as needing another look, instead of also incorrectly flagging Lateral Thinking itself.

[2.1.12] - 2026-09-18

Changed

  • Real Builder's final Build Prompt is now written by Claude Opus instead of being mechanically assembled from your answers — it reads everything decided across the whole wizard (research, the capability map, every stage's reasoning, and the accepted critique) and writes one coherent implementation brief for a coding agent, always using the strongest model regardless of which model you picked for the wizard itself.
  • If a stage was never reached (for example the wizard was ended early), the Build Prompt now says so explicitly instead of silently leaving it out.
  • If generating the Build Prompt ever fails, your previous successful one is kept — it's never lost or replaced with something broken.

[2.1.11] - 2026-09-18

Added

  • Real Builder now shows a "Show details" option on generation failures, revealing the real reason instead of just a generic message — and every failure is saved against the project so it can be reviewed later from a "Show error history" panel, not just seen in the moment.
  • Starting a new Real Builder project now lets you give it a title up front, instead of always naming it from the first line of your request.

[2.1.10] - 2026-09-18

Added

  • Your own Dashboard now uses the same at-a-glance layout as the admin trainee view: lessons and scenarios done as clickable numbers with a full list/detail view below, plus a "Suggested next steps" note pointing you toward what to do next.

Changed

  • "Awaiting review" is no longer its own separate number or tab — a mentor note was never required for a lesson or scenario to count as done. Instead, anything still waiting on a note is now flagged right inside the lessons/scenarios list itself.
  • Content that's queued but not yet served to a trainee is now shown in an unmistakable blue everywhere it appears, so it's never mistaken for something already completed.
  • Hovering a number on a dashboard or the Trainees page no longer shifts the layout — the explanation now fades in over reserved space instead of pushing everything else around.

Removed

  • Removed the "Real Problems" link from the navigation now that Real Builder covers that ground.

[2.1.9] - 2026-09-17

Changed

  • The admin Trainees page is now a clear summary (how many trainees, awaiting review, low on queue, sessions completed) plus one table, instead of a crowded row per trainee.
  • Clicking into a trainee now opens a single page with five clickable numbers — lessons done, scenarios done, awaiting review, and what's queued — each one showing the matching list and full detail (questions, answers, prompts, feedback) side by side, so it's quick to see how a trainee is progressing and whether they have enough queued to keep going.

Added

  • After editing a trainee's goal prompt, admins are now offered a one-click option to rebuild the content already queued for them so it reflects the new prompt.

[2.1.7] - 2026-09-17

Added

  • Admins can now delete a Level 1 question from a trainee's session before it's been answered — useful for removing a bad or broken question without touching the rest of the lesson.
  • Voice input, read-aloud, and a new "Adaptive Level 2" mode can now be turned on or off per trainee from the admin Edit User screen, instead of only all-or-nothing for everyone.
  • Adaptive Level 2: after finishing a Level 2 scenario, a trainee with this turned on can ask for a fresh one built specifically Easier, Tougher, or targeting a skill they've recently struggled with — generated on the spot rather than picked from a fixed queue.

[2.1.6] - 2026-09-17

Added

  • A logo for AI Prep — a twin-sparkle mark — now shows in the browser tab, as the app icon if you add it to your phone's home screen, and next to the name in the top nav.

[2.1.5] - 2026-09-17

Added

  • A footer at the bottom of every page showing the current version, a link to this Change Log, and who built it.

[2.1.4] - 2026-09-17

Changed

  • "🔊 Read aloud" now shows real progress while it's generating instead of a plain "Loading" message, can be paused and resumed instead of only stopped, and highlights each word as it's spoken.

[2.1.3] - 2026-09-17

Added

  • A new "Real Builder" section: describe a business idea in plain language and AI infers the fuller system around it, walks you through a guided set of decisions (users, workflows, design, data, security, tech stack, and more), and ends in a System Blueprint and a ready-to-use build prompt for a coding agent.
  • The AI instructions that power Real Builder now show their version and a short fingerprint right in the toolbar, with a "View Prompt" option to see exactly what's driving it — and every project permanently remembers which version built it, even after the instructions are later updated.
  • Real Builder can now research a business's own site and how others solve similar problems (commercial products, open source, academic work, adjacent industries) before proposing a system, and folds that research into its first questions instead of asking blind.
  • That research now keeps running in the background even if you close the browser or the app restarts partway through — it no longer gets silently stuck if the connection is interrupted.
  • A "🔊 Read aloud" button lets you listen instead of read: on Level 1 explanations, questions, and answers; Level 2 scenarios; and throughout Real Builder (questions, research findings, the Blueprint, and the Build Prompt).
  • A microphone toggle for speaking your prompt instead of typing it is now on every main text box in Real Builder, not just the starting one.

[2.1.2] - 2026-09-17

Fixed

  • Generating a batch of pre-seeded lessons or scenarios for a trainee could fail partway through and leave no clear record of how many actually made it in. It now always finishes and reports the real count, even if one item along the way has trouble.

[2.1.1] - 2026-09-16

Added

  • The Trainees page now has a Recent Activity feed showing what everyone has actually been doing — who started or finished a lesson, answered a question, submitted a Level 2 prompt, or worked on a real problem — each one linking straight to that piece of work, newest first.

[2.1.0] - 2026-09-16

Changed

  • Level 2 sessions now show a side panel with your attempt number, how many turns you've written, and how the difficulty compares to your last session — no more empty space next to the scenario.
  • Attaching a file on "Approach a real problem" now uses a proper button with a removable list of what you've picked, instead of the plain browser file control.

[0.1.15] - 2026-09-16

Changed

  • The whole app has been redesigned around one consistent look. Every screen now uses the same width, spacing and page title style, so pages no longer shift or resize as you move between a lesson, a session and a completed session.
  • Lessons and scenarios now read like an article — a serif reading typeface, a comfortable line length, and a side panel on Level 1 lessons showing where you are instead of empty space.
  • Level 1 and Level 2 are now visually distinct, so it's obvious at a glance which kind of work you're looking at, on the dashboard, in your history and when starting a new session.
  • Buttons, badges, scores and empty screens have been made consistent everywhere, and the main action on each screen is now clearly the most prominent one.

Fixed

  • On several screens the text ran right up against the edge of its card. It now has proper spacing.
  • Long session descriptions no longer stretch a row to several times the height of its neighbours.

[0.1.14] - 2026-09-16

Added

  • Level 1 lessons and Level 2 scenarios can now include a diagram when it genuinely helps explain something, like how a request flows through a system or the steps in a process.

[0.1.13] - 2026-09-15

Fixed

  • Pre-generating Level 1 lessons for a trainee on an Opus model almost always failed. Fixed by giving that generation step enough room to finish its work.

[0.1.12] - 2026-09-15

Added

  • Your dashboard now shows a highlighted panel comparing how much Level 1 and Level 2 work you've done, with a nudge toward whichever one you've been neglecting.
  • Admins can see each trainee's score history — Level 1 accuracy and Level 2 scores as simple charts — from the Users list.
  • Admins can generate more pre-seeded content for a trainee right from the Users list, without opening their page.

Changed

  • The "queued content" and "to review" numbers on the admin Users list are now clickable, taking you straight to that trainee's queued lessons/scenarios or the sessions waiting on a comment.

[0.1.11] - 2026-09-16

Changed

  • The admin Users list now shows each trainee's numbers (lessons and scenarios completed, concepts understood, and how much pre-seeded content is queued) as clear number tiles instead of a cramped line of text.
  • Level 2 sessions now use more of the screen, and the box you write your prompt in is noticeably taller — both make more room for longer scenarios and longer prompts.

[0.1.9] - 2026-09-15

Added

  • Admins can now pre-generate Level 1 lessons and Level 2 scenarios ahead of time for any trainee, so their next session starts instantly instead of waiting on the AI. A new "Pre-Seeded Questions and Answers" screen lets you page through everything queued — one lesson or scenario at a time — before the trainee ever sees it, and drop anything you don't want served.

Changed

  • The admin trainee-edit page now uses the full width of the screen instead of a narrow column.

Fixed

  • Level 1 sessions would occasionally fail to generate (showing a "malformed" error, or sometimes a network error) — this happened whenever the AI wrote five answer choices instead of four for one of the questions. That's now handled gracefully instead of failing the whole session.
  • Errors during session generation now show a plain, friendly message instead of an internal one.

[0.1.8] - 2026-09-14

Added

  • Level 2 scenarios now have an in-session Help button (for a nudge without giving away the answer) and a quick way to rate how clear and how hard the question was.
  • Admins can now fully reset a trainee's history (sessions, real-problem threads, and curriculum progress) from their user page, without touching their login.

Changed

  • Level 1 lessons generate noticeably faster: the explanation and the four questions are now written at the same time instead of one after the other, with an on-screen readout of how long each part is taking.
  • AI-written text (lessons, scenarios, coaching feedback, and the Real Problem advisor) now shows proper formatting — bold, bullet lists, and the like — instead of literal ** and - characters.
  • The New Session and Approach a Real Problem pages are now the same width as the pages they take you to, so the screen no longer resizes right after you submit.

Fixed

  • The Real Problem advisor could occasionally come back with a completely empty reply and no error. It now recovers automatically, and still streams in as it's written instead of appearing all at once.
  • Level 1 multiple-choice questions are harder to guess by spotting the longest answer.

[0.1.7] - 2026-09-14

Fixed

  • Voice input now shows the real reason it didn't work instead of always blaming the microphone. If your account's voice service isn't reachable, it says so instead of telling you to check microphone permissions you already granted.

[0.1.6] - 2026-09-14

Fixed

  • Removed a thin bar that kept sweeping back and forth under the top navigation whenever the app was waiting on an AI response. It was more distracting than useful.

[0.1.5] - 2026-09-14

Changed

  • Level 1 lessons now show the explanation right away instead of it appearing only after the four questions are also ready, and the progress bar moves more smoothly while the questions are being written.

[0.1.4] - 2026-09-14

Changed

  • Responses should feel faster, especially for trainees with a longer Goal Prompt set up for them — that context no longer has to be reprocessed from scratch on every request.
  • AI-written text (lesson explanations, scenarios, coaching feedback, advisor replies) now reads a little more plainly and naturally.

[0.1.3] - 2026-09-14

Changed

  • Refreshed the look of the whole app: clearer titles, a mobile-friendly layout with a bottom navigation bar on phones, and dark mode fixes.
  • Every AI response (lessons, scenarios, coaching feedback, advisor replies) now shows a real progress bar with what's happening and how long it usually takes, plus a way to stop and try again if it's taking too long.
  • The dashboard and admin pages now show your Level 1 / Level 2 progress and which concepts you've understood at a glance.
  • Level 2 scenarios now focus on practicing real AI-building skills (discovery, model choice, scoping, security, verification, and more) instead of testing how well you can work with a vague or incomplete request.

Fixed

  • Admin section tabs and navigation links now correctly highlight the page you're on.
  • The admin sessions list now shows the right level and score for completed Level 1 sessions instead of "not scored yet".

[0.1.2] - 2026-09-14

Added

  • New Level 1 "Learn" sessions: a short lesson on one concept at a time, followed by four multiple-choice questions to check your understanding. Choose Level 1 or Level 2 when starting a new session.
  • The dashboard now shows which level each session was, and your score for Level 1 quizzes.

Changed

  • Scenarios and coaching feedback now appear as they're written, instead of showing a plain "generating…" message the whole time.

[0.1.0] - 2026-09-14

Added

  • First working version of AI Prep: practice scenarios with instant coaching feedback (what you did right, what you did wrong, how to improve), session history, and an admin area for managing trainees and reviewing their sessions.