Career agent · live demo

Your expertise, redrafted for the AI era.

For freelance copywriters whose clients draft with ChatGPT now. It reads your story and notes, then gives 3 paths where your skills carry over and 3 small actions for this week.

Try the demo See how it works
Simulation · synthetic data

Built for copywriters whose clients draft with AI

Discovery interviewsPositioning and messagingBrand voiceCustomer languageEmail copyLanding page copyEditing and critiqueTeaching and workshops
Why it matters

Your clients draft with ChatGPT now. What you do best still lives in your notes, your calls and the links you save, but it is scattered, and planning the next career alone tends to stall halfway.

NOTE
Polish-job rate noteNOTE · OCT 9
CAL
Discovery call, skincareCALENDAR · 45 MIN
LINK
AI drafts sound the sameSAVED LINK · OCT 10
TXT
Career: 7 yrs of copyYOUR STORY · 3 BOXES

[SIMULATION] Sample inputs from Dana, a synthetic persona

Your notes and story go in. Out come three paths that keep your skills and three actions for this week.

Three paths that keep your skillsEvery card starts with What carries over, then What you'd need to build and How you'd start.
Three small actions this week30 minutes or less each, 10 on low-energy days. Not today moves a step, never deletes it.
Rules decide, AI only drafts the wordsEvery call comes from tested rules, and Why? shows the evidence behind it. If the AI is unavailable, a labelled sample answer stands in.
"My clients use ChatGPT now. What do I do?"
[SIMULATION] Dana, 38, freelance copywriter · synthetic persona, not a real user Walk through Dana's week

Why now: writing work is already shifting. Each number keeps its source label.

Upwork writing earnings down 5.2%Monthly, for Upwork writing freelancers after ChatGPT launched; jobs down 2%. Hui, Reshef & Zhou, 2023. Verified.
AI named in the writers' outlookBLS: AI use for writing "is projected to dampen demand" for writers. Occupational Outlook Handbook, 2025-35. Single source.
120,136 US job cuts citing AIAnnounced cuts across all industries where the employer named AI, Jan to Sep 2026. Challenger, Gray & Christmas. Single source.
How it works

From your story to this week's plan, in four steps

  • 01
    Tell your storyThree boxes or the sample. Nothing you type is stored.
  • 02
    Connect only what you allowNotes, links, calendar. Texts stay off by default. Demo data is synthetic.
  • 03
    Choose one of three pathsWhat carries over, what you'd need to build, how you'd start.
  • 04
    Walk it this weekThree small actions. Not today moves a step, never deletes.
Go to market

Free for people first. Then we sell the brain.

One decision brain, two businesses: a free app that earns people's trust while AI reshapes their work, and a paid brain API for the companies that already pay to move people into new work.

Who it's for
53.2MUS workers in jobs where AI exposure is very high

Anyone whose work AI is changing

Bring your story and your notes; get paths that keep your skills and small steps for this week. We reach people where they already work and talk, with a sample result, not ads.

Free for peopleMetric: 2-week check-in retentionBLS AI exposure data, 2025 employment (E68)
Months 0 to 9 · plan
704,300creative and language freelancers, US and Canada: the first market

Start with freelancers, grow job by job

Freelancers pay for their own transition, so they feel the gap first. Each new job adds one skill map; the rules and the app stay the same.

One skill map per jobSame brainBLS 2025 and Canada Job Bank (E25)
Months 9 to 24 · plan
$600 to $2,600what companies pay today per laid-off employee, 3-month program

Employers and outplacement buy the brain

120,136 US layoffs cited AI in 2026 so far. Companies can run the same brain for their people, paid per call. The MCP server it plugs into runs today.

Brain API, paid per callNo subscriptionIntoo, RiseSmart list prices (E70) · Challenger, Sep 2026 (E59)

Plans, not revenue: no customers or contracts yet.

Architecture

Rules make the call. Claude only writes the words.

Tested rules make every decision, in your browser or on our decision brain. Claude only words the paths, and if any step is down the demo keeps working.

01 How a click flows through the system

How a click flows through Next Chapter The browser loads the Vercel site, which shows the demo app in an iframe. The demo app calls /api/paths, a Vercel Function that makes one Claude Sonnet 5.5 call with a labelled fallback. The demo app also calls the rules-only decision brain on a team-hosted server through a Cloudflare tunnel, retries a backup tunnel when the primary fails, and falls back to on-device signals when both are down. An MCP client and the Android lifelog app are optional inputs. TEAM-HOSTED SERVER HTTPS IFRAME POST /API/PATHS MESSAGES API /V1/THINK MCP JSON (ROADMAP) USER Judge's browser laptop or phone no login WEB Vercel site landing page at / static + Functions APP Demo app in the phone React, /app.html iframe app/logic.js in browser API /api/paths Vercel Function 1 Claude call, 12 s cap labelled fallback EXT Claude Sonnet 5.5 path wording only key in Vercel env EDGE Cloudflare tunnel primary + backup CORS: Vercel origin CORE Decision brain brain/ + app/logic.js rules only, no LLM 8 MCP tools + REST OPTIONAL MCP client Claude Desktop or Code connects to /mcp OPTIONAL Android lifelog app on-device, 11 sources JSON export FALLBACK Primary tunnel down: retry the backup. Both down: on-device signals, offline line. LEGEND Our code External or user Optional Request Optional or roadmap
The purple path is the one that matters: the demo app asks the decision brain through a Cloudflare tunnel, and the brain answers with rules, not an LLM. If the primary tunnel fails the app retries the backup; if both are down the demo keeps working and the Progress tab shows on-device signals and an offline line.

02 What each layer is built with

Next Chapter technology stack Six layers: presentation, judgment, serving, AI, mobile and quality, with the technologies in each. The judgment layer, where deterministic rules make every decision, is highlighted. LAYER BUILT WITH L1 Presentation Landing page · HTML/CSS · three.js 0.160 3D phone · animated background Demo app · React 18 + TypeScript + Vite, served at /app.html L2 Judgment Rules decide, the LLM only writes sentences app/logic.js · pure deterministic functions: path ranking by skill overlap, week plan, Not today moves, condition sizing, progress signals brain/ · person, state, goal, plan, constraint, monitor, policy, explain · Node 22 · MCP SDK, 8 tools L3 Serving Vercel · static files + Node Functions (/api/paths) Cloudflare tunnel from the team server, plus a backup tunnel · REST /v1/think · MCP /mcp L4 AI Claude Sonnet 5.5 · path wording only Anthropic Messages API · structured JSON · 12 s limit, labelled fallback L5 Mobile Android app · Kotlin · Jetpack Compose · Room · WorkManager · minSdk 29 On-device lifelog, 11 sources, JSON export · optional L6 Quality vitest (web) · Playwright e2e · node --test (logic) Gradle unit tests + lint · privacy scan
Every choice the app makes comes from the deterministic Judgment layer. Claude is called only to word the suggested paths, and a labelled sample answer stands in when it is unavailable.