Cafecito AI·Apprenticeship v2.1
v2.1 · the apprentice handbook · 2026-05-11

Forward-Deployed Engineer Apprenticeship.

If a Hialeah bakery still takes wholesale orders by fax in 2026, that's not their problem. It's ours. From hire to AI blackbelt in six weeks. No slop. Ship real things to real URLs. Sell to real Miami companies.

Cafecito AI · BoostAI Search
Cohort #1 · Evan Seder · 6-week sprint
May 12 → June 22, 2026
evan-hub →
Start here · 45 min
Day 1 Setup + Prospect Sprint
Make every account, ship gap-finder, then prep Tracy Bakery and Family Medical Group for first outreach.
Open the checklist

Today only: work this path

The full handbook is here when you need it. If you are overwhelmed, ignore everything else and move through these three checkpoints.

1 · ReadyFinish setup and get gap-finder live at a real URL.
2 · AttackOpen the Tomorrow Top 5 — prep Tracy Bakery or Family Medical Group, then make the call.
3 · AdvanceImprove one sandbox demo or set one commercial next step from the loop.

Agent bootstrap: paste this site into Claude Code or Codex

This page is meant to be self-bootstrapping. Open Claude Code or Codex in a clean repo and paste the prompt below. The agent should enter planning mode, read this page, create the Cloudflare-first starter, ask for missing secrets, and get you to a deployed first URL.

Copy-paste prompt
You are helping me start the Cafecito AI forward-deployed engineer apprenticeship.

Use this handbook as source of truth:
https://cafecito-ai.com/new-hire

First enter plan mode. Create a concise plan with:
1. Lead Agent: sources and scores prospects, writes call briefs, and outputs the next 5 human calls.
2. Builder Agent: turns the best wedge into a Cloudflare-hosted wow MVP with fake data.
3. Human Sales: uses the call script to move the buyer from cold to interested to paid pilot.

Then execute the plan.

Build requirements:
1. Set up a Cloudflare-first sandbox project for the first product demo.
2. Use Cloudflare Workers + Hono as the app runtime.
3. Use D1/R2/KV only if the demo needs persistence, files, or state.
4. Use OpenAI API and Claude API as the default model providers. Do not add extra SaaS accounts unless the product truly needs voice, SMS, email, booking, or payments.
5. Create two runnable agent routes:
   - /agents/lead: accepts a company URL or niche, returns scored leads, wedge hypothesis, call opener, and next action.
   - /agents/builder: accepts a lead brief, product block, and fake data, returns a 48-hour MVP plan and the demo route to build.
6. Create /gap-finder?url= that fetches a public page and returns one observable AI/automation gap plus a suggested product block.
7. Store all keys as Wrangler secrets. Never write API keys into source.
8. Use fake data only. No PHI, payment data, private customer lists, or client production credentials.
9. Deploy to a real workers.dev URL.
10. Print the deployed URL, test commands, top 5 leads, first MVP route, and the next human sales call.

Before coding, inspect the repo and tell me the exact files you will create or edit. If Cloudflare login or secrets are missing, stop and give me the exact commands to run.
Lead AgentRuns sourcing, scoring, public evidence capture, buyer hypothesis, opener, and call queue. Output is five calls a human can make today.
Builder AgentTurns the best lead brief into a fake-data wow MVP on Cloudflare. Output is a working URL and a 60-second demo path.
Human SalesDoes the relationship work: wedge call, objection handling, pilot ask, close. The agents prepare the shot; the human takes it.
Rinse and repeatEvery call updates the lead brief. Every demo updates a product block. Every close turns into reusable delivery infrastructure.

Note for Evan Ratner: tomorrow-morning tour

This is the handoff note instead of an email. Use it to onboard the new hire and any agent you point at the system.

1. Start on this pageOpen /new-hire. Use the Today path, then the Agent bootstrap. The page explains the job: source leads, build fake-data MVPs, make human sales calls, and turn wins into reusable blocks.
2. Set the sandbox boundaryBefore the hire builds anything live, confirm Cloudflare access, OpenAI API, Claude API, Wrangler secrets, fake data, spend caps, and no PHI/payment/customer production data.
3. Use the tool beltIn §08, pick a block by job-to-be-done: starter wedge, revenue recovery, ops cleanup, voice/SMS, trust/compliance, or delivery accelerator. Do not let the hire invent a custom one-off first.
4. Use the prospect mapOpen /new-hire/prospects#map. It is the source of truth for the 21-prospect mapping: primary block, secondary block, bundle, demo path, and pilot ask.
5. Run the two-agent loopLead Agent produces the top five call queue and briefs. Builder Agent ships one Cloudflare-hosted fake-data wow path. Human seller makes the call and logs exact buyer language.
6. End each day with artifactsOne lead brief updated, one demo improved or shipped, one sales call or followup logged, one reusable asset sharpened, and one next commercial action scheduled.
If you have 30 minutes
  1. Open Today only and pick your current checkpoint.
  2. Open §02 · Setup if the checklist is not complete.
  3. Open §09 · Hot List if you need a prospect to attack.
  4. Open §10 · Loop if you are deciding what to do next.

First 48 hours: what done means

This is the operating checklist. By the end of Day 2, there should be artifacts, not notes about intent.

1. Worker livegap-finder-<name> deployed at a real Workers URL with one prospect URL already tested.
2. Two call briefsTracy Bakery and Family Medical Group each have a one-page brief: wedge, questions, opener, demo promise.
3. First call block loggedAt least 2 calls attempted, with outcome, exact words, voicemail/live status, and next step in the hub.
4. One demo spec chosenPick the highest-pain response and write a 48-hour v0 build plan mapped to a product block.

Evan R sandbox/API setup

The intern can build the demo, but Evan owns the sandbox boundary. No prospect demo should use production credentials, real PHI, real payment data, or uncontrolled SMS/voice spend.

1. Cloudflare sandboxCreate or confirm a demo Worker namespace, D1 database, R2 bucket, and short route like demo.cafecito-ai.com/<prospect>.
2. API keysProvision limited Anthropic/OpenAI, Twilio/VAPI, Resend, Cal.com, and Stripe test keys as needed. Put them in Wrangler secrets, never source.
3. Demo data onlySeed fake orders, fake patients, fake calls, fake customers, and fake quote requests. If the prospect sends real data, stop and scope a paid pilot.
4. Spend guardrailsSet per-demo caps: max calls, max SMS, max model spend, max email sends. Log every external API call.
5. Client-wow routeEach demo gets a short URL, optional password, a reset button, and a preloaded scenario that works in under 60 seconds.
6. Pilot handoffWhen a prospect says yes, Evan converts sandbox to pilot: contract, scoped data access, billing, compliance review, and integration plan.
Full handbook · open only what you need
§ 01 Why you're here.

This is not a course. It is a forward-deployed engineering apprenticeship. You will live inside real Miami SMB problems, build software that runs in production, and earn the right to sell it.

By the end of week six you have found a Miami mid-market company, gotten a decision-maker on the phone, and closed them on a paid engagement. You have shipped four to six deployed mini-products with real URLs. You have led a teach-up that taught your mentor something they didn't know.

Your primary client engagement is Faceoff (Bartlett Hockey) — a chief-of-staff dashboard for Scott Bartlett's hockey agency. Everything else — the calling motion, the mini-products, the teach-ups — is sharpening tools to serve that client and find the next one.

The motion is find → call → close, not "post on LinkedIn → wait for inbound." The phone is the primary tool. The agent we build, the prospect list we research, the prototypes we ship — they all exist to make the next call land.

Every line of code, every email, every call has a specific recipient and a specific reason. If it could have been generated by anyone for anyone, throw it away.

You are not here to produce slop. You are here to be the AI team a Miami company didn't know they needed.

§ 02 Set up your arsenal.

Get this done before your first standup. Check off as you go — this list saves to your browser. Treat it as doctrine: ship the hello-world Worker on Day 1 or the apprenticeship hasn't started.

Setup progress0 / 0
  • Hardware & environment
  • Core accounts (one workspace first)
  • Adapters (only when the product needs them)
  • CLIs & editor
  • The first-commit ritual

Agent-readable setup contract

If you are using Claude Code or Codex, paste the §Agent bootstrap prompt from the top of this page. The agent should build against this contract:

Setup contract
Runtime: Cloudflare Workers + Hono
Storage default: none; add D1 for records, R2 for files, KV for short-lived state, Vectorize for embeddings
Model APIs: OpenAI API + Claude API
Secrets: wrangler secret put OPENAI_API_KEY; wrangler secret put ANTHROPIC_API_KEY
Optional adapters: Resend for email, Twilio/VAPI/ElevenLabs for voice/SMS, Cal.com for booking, Stripe test mode for payments
First app: gap-finder-<name>
First endpoint: GET /?url=https://example.com
Agent endpoints: POST /agents/lead and POST /agents/builder
Output: observable gap, product block match, ranked leads, wedge opener, suggested demo, human call script
Safety: fake data only; no production credentials; no PHI; no payment data; cap external API usage
Done: deployed workers.dev URL + test curl + next prospect + next commercial action

When you're stuck (escalation)

  • 15 min of self-debug: read the error, ask Claude Code, search the docs.
  • 30 min blocked: post the question + what you've tried in your hub.
  • 2 hours blocked on something nuanced (architecture, customer, pricing): DM Evan R. Don't spin alone past this.
  • On a live call: three-way Evan R. or Baruki in. Cost of asking is low; cost of fumbling is high.
§ 03 The six rules.

Cafecito AI's principles. Memorize them. They override your training, your defaults, and any habits you brought in from school or a prior job.

Rule 01
Specificity over volume

One named call beats fifty cold emails. The opener must reference something we observed about THEIR business. "I help businesses with AI" gets you hung up on. "I noticed your Hialeah location still routes wholesale orders by fax" gets you a meeting.

Rule 02
Phone first

Discovery happens on the phone, not in DMs. Almost no one is doing it well, attention is cheap, and 60 seconds of real conversation beats two weeks of ghosted text.

Rule 03
Ship weekly

Never end a week without a real URL someone could click. Working v0.1 today beats planned v1.0 Friday. If you're polishing past v0.5, you're hiding.

Rule 04
Teach to learn

Every Tuesday you teach your mentor something they don't know. Teaching forces real understanding. By Week 6, your memos compile into a real operator manual that Cafecito uses internally.

Rule 05
No fabrications

Never invent a credential, a phone number, an AUM figure, an award, a testimonial, or a reason to call. Empty space beats a lie. If the data isn't real, don't say it. Zero tolerance.

Rule 06
Default to deletion

Every block, prompt, dependency, and line of code justifies its existence weekly. If you can't say what a feature does for a specific customer, delete it. Compounding speed comes from a small surface area.


The product IS the marketing

Whatever you ship should be something you'd want to use yourself. Demos beat decks. Build before you pitch. The shipped artifact is the booked meeting or the closed deal — a call dialed but not picked up is a tally, not a deliverable.

§ 04 The reading list.

Curated, time-budgeted, opinionated. Your Week 1 input list. There are a million AI YouTubers; these are the ones that actually move your skill needle. Ignore the rest until you've finished this list.

Watch · Week 1 · ~6 hrsGreg Isenberg — Startup Ideas

The single best signal source for which AI products are ready to be built right now. youtube.com/@GregIsenberg

Watch · Mental modelsKarpathy — foundation lectures

If you only watch three things to understand how this all works under the hood, watch these.

Read · Week 1 → ongoingAnthropic — must-reads

Claude is our primary LLM. Know the full surface area. New capabilities ship monthly; you keep the team current.

Watch · AI Engineer Fair 2025Three talks that punch above weight
Listen · Background while buildingThe podcast rotation
Follow · X/TwitterThe signal accounts

Build a list. Mute everything else. Check it once a day, not all day.

  • @karpathy — Mental models for LLMs. OG teacher.
  • @AnthropicAI — Official source for Claude / MCP / Skills launches.
  • @alexalbert__ — Anthropic DevRel; first to demo new Claude features.
  • @simonw — Sharpest practical-LLM blog.
  • @gregisenberg — Distribution playbooks, vertical-AI ideas.
  • @swyx — Latent Space / AI Engineer Fair.
  • @levelsio — Solo-dev SaaS playbook.
  • @rauchg — Vercel; v0 + AI-native frontend.
  • @ideabrowser — Daily validated startup ideas.
  • @dexhorthy — Context-engineering frontier; agent reliability.
Read · Stack fundamentals · ~3 hrsCloudflare + Hono + Stripe + prompting

You can't ship the §08 product blocks without these. Fast, hands-on, official docs only.

Tools · Use Day 1ideabrowser.com

Greg's productized idea database. Validated startup ideas with full market reports. MCP integration wires idea context into Claude Code.

  • Idea of the Day — free daily validated idea.
  • Tools page — MCP search API. Configure in Claude Code.
  • Pricing — free tier is enough for the apprenticeship.

Belt drills · Weeks 1–6

  • Week 1: gap-finder Worker live + Day-1 setup complete + first 5 cold calls dialed.
  • Week 2: Ship SKILL.md codifying your prospect-research workflow. 8–10 calls dialed.
  • Week 3: Ship MCP server exposing Sunbiz officer-lookup. First 20-min discovery call solo.
  • Week 4: Ship a §08 mini-product to a real URL with Stripe. First commercial proposal sent.
  • Week 5: Verbal-yes on a paid pilot. Or shipped a Faceoff feature unsupervised.
  • Week 6: Close. Sign. Synthesize the operator manual. Onboard cohort #2.
§ 05 Cloudflare-native by default.

This is the canonical Cafecito stack. Build everything in one place by default: Cloudflare for runtime/data, OpenAI API + Claude API for model calls, GitHub for source. Do not add infra we don't need. Tempted to introduce Vercel, Supabase, AWS, Postgres, a standalone database, or another SaaS dashboard? Ask first.

Default build policy
  1. Start on Cloudflare: Worker + Hono first, then D1/R2/KV/Vectorize only when needed.
  2. Use both model APIs: Claude for reasoning, coding, extraction, and agentic workflows; OpenAI for GPT-backed product features, evals, and fallback model coverage.
  3. Adapters are optional: Resend, Twilio, VAPI, Cal.com, Stripe, Apollo, Clay, and Yelp are added only when the product cannot wow the client without them.
  4. Secrets live in Wrangler: every API key goes through wrangler secret put. No keys in source, screenshots, prompts, or demo data.
Compute
  • Workers · Hono.js
  • Workers AI · cron jobs
  • Pages · Workflows · Queues
  • Durable Objects when needed
Storage
  • D1 SQL
  • R2 object storage
  • KV key-value
  • Vectorize for embeddings
LLM
  • Claude API reasoning + extraction
  • OpenAI API GPT product features + evals
  • Claude Code + Codex for build agents
  • Choose model per demo, not per habit
Agents
  • Claude Code and Codex
  • MCP servers
  • Agent SDKs where useful
  • gstack skill library (50+)
Web
  • Hono server
  • Vanilla HTML + Tailwind CDN
  • Astro for content
  • htmx where it fits
Optional adapters
  • Stripe only for payment demos
  • Resend only for email flows
  • VAPI/Twilio only for voice/SMS
  • Cal.com only for booking
Research & data
  • Firecrawl web scraping
  • Sunbiz (FL corps)
  • Apollo · Clay (outbound)
  • Yelp/Google APIs (reviews)
Auth
  • Cloudflare Access internal
  • HMAC tokens for portals
  • OAuth where needed
  • No sessionStorage in artifacts
§ 06 The cadence.

Three nested rhythms. Daily, weekly, sprint. Don't break the cadence — momentum is the moat.

Daily

  • 9:00 AM — standup ticket sync in your hub. What shipped yesterday, what's blocking, what you'll touch today.
  • EOD — 60-second Loom recap of what you shipped. Posted to your hub. Yes, every day.

Daily Client-Closing Loop

Do not wait for Wednesday to call. Do not wait for Thursday to build. Every day touches lead, wedge, prototype, and sale until a client closes.

ComponentDaily output
SourceAdd or refresh 2-3 leads. Every new idea, referral, weird workflow, and prospect goes into the hub.
ScorePick the best 1-2 leads to attack today using pain, demo speed, buyer access, reuse, and sandbox readiness.
BriefCreate or update one call brief: public wedge, buyer hypothesis, opener, questions, demo promise, pilot ask.
CallMake 2-5 outbound attempts or followups. Log live/voicemail/no-answer and exact buyer language immediately.
PrototypeAdvance one sandbox demo. It can be a new route, stronger fake data, better prompt, cleaner output, or reset path.
SellAsk for the next commercial step: discovery, demo, paid pilot, proposal, or close. Do not end with vague "send info."
ExtractUpdate one reusable asset: product block, prompt, call script, objection response, demo checklist, or prospect dossier.

Daily done criteria

2+leads touched
2+outbound attempts
1demo advanced
1next action set

If a day has no lead touched, no outbound attempt, no prototype movement, and no next action, it did not move the apprenticeship toward a closed client.

Closed client mode

Once a client closes, delivery becomes the primary daily block, but the loop does not stop. The split changes: 70% build/deliver the closed client project, 20% active followups and demos, 10% new lead sourcing. Delivery work must still extract reusable product-block improvements.

Daily log template

Copy into the hub before EOD
Date:
Leads touched:
Outbound attempts:
Live conversations:
Exact pain quote:
Prototype/demo advanced:
Demo URL:
Commercial ask made:
Followups sent:
Closed-client delivery milestone, if any:
Tomorrow's next action:

Standing weekly anchors

AnchorPurpose
Evan R standupReview current lead queue, sandbox readiness, prototype quality, and closed-client delivery risk.
Baruki call blockTighten openers, run live outbound, sharpen objection handling, and push prospects to demo/pilot.
Teach-upShare one thing learned from building, calls, customer language, or a new AI workflow.
RetroExtract reusable assets and decide what to keep, kill, or turn into the next product block.

Sprint

Six weeks. May 12 → June 22. The belt test (§11) tells you if you got there. The hub at cafecito-ai.com/evan/ is the source of truth for what you're working on right now — projects, tickets, notes. Update it daily.

§ 07 The work that exists.

Real client engagements ranked by what we'd hand a hire today. You won't touch all of these — Faceoff is your primary line. But know what the portfolio looks like.

Primary · live · client
Faceoff — Bartlett Hockey

Chief-of-staff dashboard for Scott Bartlett's hockey agency at faceoff.cafecito-ai.com. Recruiting funnel, agent profiles, ~5k LOC. Your primary client engagement this summer.

Live · bilingual · voice
Suite Air — Miami HVAC

Bilingual HVAC for HOA/condo Miami. Vapi voice agent (Lucia) handles inbound, books service, sends Resend confirmations. Expansion: more verticals, more languages.

Live · PM portal
Lavin Eviction

Eviction case management for Danny Lavin's firm + property managers. Three real cases live; Category Five PM is the anchor. FL-franchise expansion in scope.

Live · automation backbone
Cafecito intake

Inbound-lead automation across all Cafecito products. Forms, routing, Resend, admin dashboard.

Live · MCA · white-label
eFin Cap / AC Group

MCA broker submission pipeline. Intake → parsing → credit → rules → UW → DocuSign → stips → funding. AC Group + Lexington Recovery white-label active.

Live · own SaaS
BoostAI Search

AI-visibility scanner. Sister brand. Audit a domain, get a citability score across ChatGPT/Claude/Perplexity/Gemini, 14-day email followup. Sellable today as $99 audit + $499/mo monitoring.

Live · LCS internal
Levin Capital tooling

Soft-dollar dashboard, futures advisory, portfolio intelligence portals, morning notes. Sister-firm internal use only.

By Week 3, you scope and ship one mini-product against a Miami prospect from the hot list below. By Week 5, you've sold it.

§ 08 The arsenal — 16 blocks.

Don't build one-offs. Pick a product block off the shelf, customize for one prospect, deploy in 3–5 days, charge for it. Same block redeploys to the next client. Compute is cheap. Reuse compounds. This is how a one-person Cafecito sales engineer outproduces an agency.

Tool belt: pull the right block for the job

The blocks are not a reading list. They are a tool belt. Start with the smallest wedge that gets the buyer to say, "show me that with our workflow," then bundle only after the pain is proven.

Starter wedge tools

Cheap, fast, foot-in-the-door. Closes in one call and sets up the bigger sale.

Revenue recovery tools

Pays for itself the first month. Easiest ROI math on the call.

Ops cleanup tools

Fixes internal mess that keeps owners up at night. Higher-priced, stickier.

Voice / SMS tools

Phone-based products. Highest "wow" on the live demo. Watch the usage caps.

Trust / compliance tools

Sells to risk-averse buyers. Citations, defensibility, audit trail.

Delivery accelerators

Productized, high-velocity work. Strong fit for eat-your-own-dogfood demos.

Bundle playbooks

BundleUse whenBlocksFirst demo
B01 · Bakery / Wholesale Bundle →Orders arrive by fax, email, PDFs, photos, or generic contact form.11 + 02 + 14Fake wholesale order becomes clean ops packet with missing-field flags.
B02 · Clinic / Medspa Bundle →Phone-heavy business, after-hours shoppers, bilingual demand, or front-desk risk.01 + 12 + 02Bilingual voice + missed-call SMS recovery + smart intake routing.
B03 · Contractor / Sign-Shop Bundle →Generic form cannot capture specs, photos, materials, urgency, or install details.03 + 11 + 14 + 16Photo-to-quote in 90 sec + status portal that kills "where is my job?" calls.
B04 · Restaurant / Local Services Bundle →Reviews, stale customer DB, missed calls leak revenue across all three channels.04 + 07 + 12Daily review responses + Tuesday SMS reactivation + missed-call recovery.
B05 · Solo Professional Bundle →Legal, CPA, mortgage, or solo professional needs site + reviews + intake from day 1.10 + 04 + 02Bookable site + review responses + bilingual intake form, in 3 days.
How the loop uses the tool belt
  1. Lead Agent finds a company and assigns a primary block plus one expansion block.
  2. Builder Agent builds the smallest fake-data wow path using the primary block only.
  3. Human Sales makes the wedge call and uses the demo promise to book discovery or pilot.
  4. After the call update the prospect map, sharpen the block, and decide whether to bundle.
Block 01

Bilingual Voice Receptionist

24/7 inbound call answering. ES/EN auto-detect. Books appointments. Routes complex calls.

Deploy: 2d$1.5k setup$400/mo + per-min
Stack: VAPI · Claude Sonnet · Cal.com · Resend · Twilio
Best fit: any Miami SMB with a phone ringing during business hours. ~70% of the hot list. Reference: Suite Air (Lucia), Nightline.
Block 02

Smart Intake + Triage Form

Public form → AI extracts urgency, type, dollar value → routes to the right human → SMS/email confirmation in customer's language.

Deploy: 2d$2k setup$300/mo
Stack: CF Worker · D1 · Hono · Claude tool-use · Resend · Twilio
Best fit: law firms, PM cos, MCA brokers, contractors, wholesale food. Reference: Lavin Eviction, Cafecito intake, BoostAI.
Block 03

Quote / Estimate Generator

Customer uploads a photo or fills a form. Claude Vision reads it. Generates labor + materials + total. PDF or DocuSign out.

Deploy: 3d$3k setup$500/mo
Stack: CF Worker · R2 · Claude Vision · PDF gen · Resend
Best fit: HVAC, roofing, painting, pool service, marine repair, signs, custom millwork.
Block 04

Review Response Bot

Pulls Google + Yelp reviews via API. Drafts personalized responses in the owner's voice. Owner one-click approves.

Deploy: 2d$1k setup$250/mo
Stack: CF Worker · D1 · Google/Yelp APIs · Claude · Resend daily digest
Best fit: restaurants, salons, dental, MedSpa, gyms — anywhere review velocity matters.
Block 05

AI Visibility / Citability Scanner

Scans how a brand surfaces in ChatGPT / Claude / Perplexity / Gemini. Score + email report + 14-day followup sequence.

Deploy: existing$99 audit$499/mo monitoring
Stack: existing boost-ai worker — sellable today.
Best fit: any service business discovered via search. SEO firms, law firms, healthcare, recruiting.
Block 06

Document AI Search

Upload contracts, manuals, policies. Semantic search + Q&A in employees' native language. Karpathy-PKM pattern, but for one client's corpus.

Deploy: 3d$5k setup$750/mo
Stack: CF Worker · R2 · Vectorize · Claude
Best fit: law firms (precedent search), accountants (tax memos), HOAs (bylaws), large dental/medical groups (protocols).
Block 07

SMS Reactivation Agent

Customers who haven't visited in N days get a personalized SMS. AI handles the reply thread. Books or escalates.

Deploy: 2d$1.5k setup$300/mo + SMS
Stack: CF Worker · D1 · Twilio · Claude · Cron Triggers
Best fit: salons, gyms, restaurants, dental, primary care, MedSpa.
Block 08

Cold Outreach Engine

UCC / Sunbiz / Apollo data → Claude writes personalized first message → sequence → calendar. Self-use first: this is what Evan uses on the hot list.

Deploy: 3d$4k setup$1k/mo
Stack: CF Worker · D1 · Apollo/Sunbiz · Claude · Resend · Cal.com
Best fit: brokers, agencies, B2B services. Eat your own dog food.
Block 09

Internal Ops Dashboard

Custom dashboard pulling client's existing data (QuickBooks, Toast, Mindbody, Salesforce, LiquidityBook) into one screen with AI summaries + alerts.

Deploy: 5d$7.5k setup$1.5k/mo
Stack: CF Worker · D1 · Claude · integration APIs
Best fit: multi-location operators with 3+ tools. PM firms, restaurant groups, multi-loc dental, RIAs.
Block 10

Personal Brand Site + Booking

For solo-operators / small practices. AI-written copy, online booking, payment, content engine. The cheapest entry-point block.

Deploy: 1d$2k flator $200/mo
Stack: Astro · CF Pages · Stripe · Cal.com · Resend
Best fit: lawyers, dentists, financial advisors, contractors going solo. Foot-in-the-door before upselling Block 02 or 06.
Block 11

AI Order Desk

Fax, email, photo, PDF, or form submission → normalized order packet → missing-info flags → ops email.

Deploy: 2d$2.5k setup$500/mo
Stack: CF Worker · R2 · Claude Vision · D1 · Resend
Best fit: wholesale food, distributors, specialty manufacturers. First target: Tracy Bakery.
Block 12

Missed Call Recovery

Every missed call triggers an immediate SMS. AI classifies the reply, books or routes it, and writes the staff summary.

Deploy: 2d$1.5k setup$350/mo + SMS
Stack: Twilio · CF Worker · D1 · Claude · Resend
Best fit: medspas, car wash, dental, healthcare, property management. Entry wedge before full voice agent.
Block 13

Bilingual Conversion Layer

Spanish/English overlay for an existing site: intake, FAQ, booking CTA, lead capture, and handoff email.

Deploy: 1d$1.5k setup$250/mo
Stack: CF Worker snippet · Claude · Resend · optional Twilio
Best fit: Miami businesses with English-only sites in Spanish-heavy markets. Fastest "wow" demo.
Block 14

Customer Status Portal

"Where is my order/project/request?" portal with searchable status, upload trail, staff notes, and AI-written customer updates.

Deploy: 4d$5k setup$900/mo
Stack: CF Worker · D1 · R2 · Resend · Claude summaries
Best fit: sign shops, custom millwork, property managers, repair businesses. Great for Signscenter and Saul Signs.
Block 15

AI Front Desk QA

Sample calls or intake logs → classify missed revenue, unanswered questions, language mismatch, and followup gaps.

Deploy: 2d$1k audit$300/mo
Stack: call exports · R2 · Claude · weekly Resend report
Best fit: teams not ready to replace the phone flow. Sells the pain before Block 01.
Block 16

Vertical Quote Form Rebuilder

Replace generic contact forms with vertical-specific quote flows that ask the three questions a human estimator would ask first.

Deploy: 2d$2.5k setup$400/mo
Stack: Hono · D1 · Claude tool-use · Resend · optional PDF output
Best fit: insurance, signs, car wash detail packages, contractors, property management. Strong fit for PTL and Aventura's Finest.

How to use the arsenal

When a prospect says yes to a discovery call, you arrive having already matched their observable gap to one or two blocks. The dossier in §09 names the matched block for each company. You're not selling a custom build — you're selling speed-to-deployment of a pattern that's already running for someone else.

Signal you observeLead withDemo shape
Fax, voicemail, or generic contact form handling revenue-critical requestsBlock 02 + Block 03A smart intake form that turns messy input into a routed quote packet.
High call volume, broken portal, multi-location scheduling frictionBlock 01 + Block 02A bilingual phone/intake triage flow that books, routes, or escalates.
Lots of policies, forms, PDFs, leases, contracts, or internal playbooksBlock 06Document search over their public forms or one sample PDF bundle.
Search-driven service business with weak AI visibilityBlock 05BoostAI audit report with one fix they can understand immediately.
Fax/email order desk, missed calls, English-only Miami site, or generic quote formBlocks 11-16A fake-data demo that shows the exact before/after workflow in under 60 seconds.

When you ship a new client engagement, ask: can the next client reuse this? If yes, extract it into a new block (Block 11+) and add it here.

§ 09 Miami hunting ground.

Real, named, currently-operating Miami-area mid-market companies with observable AI/automation gaps verifiable on their own public web presence. No fabrications. Every claim is linkable. The full list is useful, but Day 1 is about momentum: prep one or two companies, make the calls, log what happened.

Full hot list · 21 dossiers · 16 blocks
Open the prospect → block map
Start with the mapped board: primary block, secondary block, bundle, demo path, pilot ask, and then open the full dossier only when needed.
Open map

Tomorrow sprint: attack 1-2 prospects

Do not read all 21 dossiers first. Start with two high-signal companies, produce a one-page call brief for each, make the calls, and convert any live conversation into a 20-minute discovery.

Priority 01 · wholesale orders

Tracy Bakery

Why first: published fax number for wholesale ordering is a concrete automation wedge. The build is easy to demo without private data.

Lead product: Block 11 AI Order Desk, then Block 13 Bilingual Layer and Block 14 Status Portal if the order flow pain is real.

Day-1 goal: book a discovery on wholesale order intake, not sell a finished system.

Priority 02 · healthcare routing

Family Medical Group

Why second: 15-location primary-care footprint with patient portal friction and one-phone routing pressure.

Lead product: Block 12 Missed Call Recovery first, then Block 01 Voice and Block 15 Front Desk QA after the call pattern is proven.

Day-1 goal: learn whether calls, forms, or portal handoffs are creating the most pain.

90-minute execution block

  1. 0-15 min: open the dossier, company site, Google Maps, LinkedIn, and Sunbiz. Capture only facts you can cite.
  2. 15-35 min: run the research prompt below. Save the output as YYYY-MM-DD-company-call-brief.md.
  3. 35-55 min: create the demo spec: one screen, one workflow, one value claim. No production build yet.
  4. 55-75 min: call. If voicemail, leave a 15-second message and log it. If live, ask for a 20-minute discovery.
  5. 75-90 min: log outcome, exact words, next step, and whether the product block changed.

Top-five attack cards

01 · Tracy Bakery · Block 11

AI Order Desk

Wedge: wholesale orders still route through phone/fax/contact form.

Wow demo: upload a fake fax/photo order and watch it become a clean production order packet with missing-info flags.

Close: "If this saves one messy order handoff per day, the pilot pays for itself. Want the fake-data demo tomorrow?"

02 · Family Medical Group · Blocks 01/12

Bilingual Triage + Missed Call Recovery

Wedge: multi-location patient access, broken portal anchor, shared phone routing.

Wow demo: caller says the issue in English or Spanish; agent classifies it, routes it, and drafts staff notes.

Close: "Let's first recover missed calls before touching scheduling integration."

03 · Med Aesthetics Miami · Blocks 12/13

After-Hours Promo Capture

Wedge: aesthetics buyers shop after hours, but the promo CTA sends them to business-hours phone flow.

Wow demo: Spanish/English after-hours SMS flow qualifies service interest and books a callback.

Close: "One recovered consult can cover the pilot. Want to see it on your current promo?"

04 · Aventura's Finest · Blocks 16/12

Instant Detail Booking + SMS Confirm

Wedge: site promises 24-hour response time for inquiry handling.

Wow demo: customer picks service, car type, urgency, and availability; AI returns a quote-band and confirmation text.

Close: "Same-minute booking beats 24-hour response. Want the one-screen demo?"

05 · Laundromart · Blocks 07/12/13

SMS Loyalty + Bilingual Support

Wedge: 11 locations, one phone number, no SMS or bilingual conversion flow.

Wow demo: customer texts "pickup" or "lavado" and receives location-aware pickup/dropoff instructions.

Close: "We can start as loyalty/reactivation before touching POS."

Prompt pack

1. Prospect call brief prompt
Use only facts I provide or facts you can verify from public sources. Do not invent names, phone numbers, revenue, employee counts, or pain.

Company:
Website:
Prospect dossier notes:
Observable gap:
Matched Cafecito product block:

Create a one-page outbound call brief with:
1. Why this company is worth calling now.
2. The exact public evidence behind the wedge.
3. The likely operator pain, clearly labeled as hypothesis.
4. Three discovery questions.
5. A 20-second opener.
6. A voicemail script.
7. The smallest demo we could build in 48 hours.
8. What to log after the call.
2. Product demo spec prompt
You are helping Cafecito AI build a fast v0 demo for one prospect.

Prospect:
Observed gap:
Target user:
Lead product block:
Discovery question we need answered:

Write a v0 demo spec:
1. One-sentence promise.
2. Input fields or trigger event.
3. AI step-by-step behavior.
4. Human handoff rule.
5. Output shown to the prospect.
6. Data we must not collect yet.
7. 48-hour build plan using Cloudflare Worker, Hono, D1/R2 if needed, Claude, Twilio/Resend only if needed.
8. Price hypothesis.
Tracy Bakery opener

"Hey, this is Evan with Cafecito AI. I saw wholesale orders still route through phone/fax/contact forms. Quick question: when a grocery or foodservice customer sends in a messy order, who turns that into something production can act on? We build small AI intake systems for exactly that handoff. Worth 20 minutes this week to see if there is a fit?"

Family Medical Group opener

"Hey, this is Evan with Cafecito AI. I was looking at your locations and patient access flow. Quick question: what creates more friction right now, phone routing, portal issues, or intake forms? We build bilingual AI triage that routes patients without replacing your staff. Worth 20 minutes to compare notes?"

Deep wedge call script

The goal is to move fast from cold interruption to specific pain to a demo they can picture. Do not pitch every block. Pick the one wedge you observed, ask short questions, and make the next step feel obvious.

Cold → interested → "show me" ladder
  1. Permission: "I know I'm catching you cold. Give me 20 seconds, and if it's not relevant I'll disappear."
  2. Observed wedge: "I noticed [specific public fact: fax orders / portal friction / 24-hour response / no Spanish intake]. We build small AI workflows that remove exactly that bottleneck."
  3. Role question: "Who owns that process today when it breaks or backs up?"
  4. Volume question: "Roughly how many of those requests hit you in a normal week?"
  5. Cost question: "When one gets missed, delayed, or routed wrong, what actually happens? Lost sale, angry patient, extra staff time, rework?"
  6. Mirror back: "So the issue is not 'AI.' It is [repeat their exact words] causing [money/time/customer impact]. Did I get that right?"
  7. Demo pivot: "If I showed you a tiny version that takes [messy input] and returns [clean routed output] in under a minute, would that be worth looking at?"
  8. Paid-pilot close: "If the demo matches the workflow, the first pilot is simple: 48-hour prototype, fake-data demo first, then a small live pilot. Setup is usually $1.5k-$3k depending on routing. Should we map it for 20 minutes tomorrow?"
If they sayAnswerNext move
"We already have software.""Perfect. We usually sit in front of the system or between inbox and staff. Which part still creates manual cleanup?"Ask for the handoff they still hate.
"Send me info.""Happy to. So I don't send generic fluff: is the pain more missed requests, slow response, or staff retyping?"Force one qualifying answer before email.
"No budget.""Makes sense. If this saved five staff-hours a week or recovered one missed order, would that change the math?"Quantify before retreating.
"Can it integrate with our system?""Usually yes, but I would not start there. First we prove the intake/routing logic with fake data, then decide whether integration is worth it."Sell low-risk pilot, not big integration.
"Sounds interesting.""Good. Let's not leave it vague. I can bring a one-screen demo using your public workflow. Tomorrow or Wednesday better?"Close for calendar.
3. Wedge-call coach prompt
Act as a B2B founder-sales call coach. I am calling this prospect cold.

Prospect:
Observed public wedge:
Likely buyer:
Product block:
My draft opener:

Rewrite the call path so it can move from cold call to interested buyer quickly:
1. 20-second opener.
2. Five questions that deepen the pain without sounding scripted.
3. The exact moment to stop discovery and offer a demo.
4. A one-sentence "wow" demo promise.
5. A paid-pilot close.
6. Three likely objections and tight replies.
Use plain spoken language. No hype. No unverifiable claims.

Build recipes to have ready

Tracy Bakery v0: wholesale order intake
  1. Create a single public form: customer, store, delivery date, items, quantities, notes, photo/PDF upload optional.
  2. Claude normalizes the messy request into line items and flags missing info.
  3. Worker stores request in D1 and sends a clean email to operations.
  4. Demo with fake order data only. Do not imply we integrated with their production systems.
  5. Success metric: time from inbound order to clean internal order packet.
Family Medical Group v0: bilingual triage intake
  1. Create a bilingual intake flow: reason for visit, location preference, urgency, insurance status, callback number.
  2. Claude classifies: appointment request, urgent escalation, records question, billing, portal help, other.
  3. Worker routes to the correct queue and drafts a staff-facing summary.
  4. Demo as a web form first; voice receptionist comes after discovery validates call volume.
  5. Success metric: fewer misrouted calls and faster callback triage.

Hunting criteria

  • $5M–$50M revenue band (or 20–300 employees as proxy)
  • Multi-location or 10+ employees — not 1–2 person shops
  • Owner / operator reachable by phone
  • Observable AI/automation gap verifiable on the public web
  • Vertical maps cleanly to one of the 16 product blocks (§08)
  • Skip: Fortune 500 subs, BigLaw, national chains, pre-revenue startups

Verticals — Cohort #1 · 21 prospects

  • Specialty manufacturing — sign shops, custom millwork, Hispanic food (5)
  • Multi-location operators — laundromat chains, hand-detail car wash chains (2)
  • Professional services — litigation boutiques, estate firms, CPAs, insurance brokers (4)
  • Healthcare — multi-location primary care, dental groups, MedSpa chains (3)
  • Real estate adjacent — property mgmt, HOA mgmt, mortgage brokerages (4)
  • Hospitality / F&B — restaurant groups, boutique hotel groups (3)

Strongest call-now signals

  1. Tracy Bakery — active fax number for wholesale orders in 2026. First call because the demo is concrete and low-risk.
  2. Family Medical Group — broken patient-portal anchor + 15 locations on one phone. Second call because routing pain is likely expensive.
  3. Med Aesthetics Miami — English-only in Spanish-dominant markets. Keep as backup if healthcare access is the stronger story.
  4. Aventura's Finest — site explicitly states "24-hour response time." Good for Block 02 intake and SMS followup.
  5. Laundromart — zero digital infrastructure for 11 locations in Hispanic-majority neighborhoods. Good for Block 01/07 after the first two are logged.

How to attack the list

  1. Run gap-finder Worker on each company URL before calling. Output: 1-page call brief.
  2. Mon/Wed/Fri = 5 calls each. 90-minute call blocks. Calls in batches.
  3. Capture the prospect's exact words describing pain. These become marketing copy + opener phrases.
  4. Day 1 calls aim for "qualify + book a 20-min discovery." Not "close." Don't jump.
  5. Log every call in your hub. Date, contact, length, outcome, exact quote, next step.
  6. Add new prospects as you find them — open a PR against prospects-html.js. The list compounds.
★ TODAY Tomorrow command center.

This is the morning operating board. The canonical 21-prospect map lives at /new-hire/prospects#map. This section exists to keep the hire from wandering: five calls, one demo, one logged next action.

Call 01 · Tracy Bakery · Block 11

AI Order Desk

Human call: ask who turns fax/email/contact-form wholesale orders into production packets.

Builder task: fake fax/PDF order upload → extracted SKUs → missing delivery date → ops packet.

Win: 20-minute discovery on wholesale order cleanup.

Call 02 · Family Medical Group · Block 12

Missed Call Recovery

Human call: ask whether phone routing, portal issues, or intake forms create the most patient-access friction.

Builder task: fake missed call → bilingual reply → location routing → staff summary.

Win: discovery on one-location recovery pilot before touching voice automation.

Call 03 · Med Aesthetics Miami · Block 13

Bilingual Conversion Layer

Human call: ask what happens to Spanish-speaking buyers who hit the promo CTA after hours.

Builder task: Spanish promo page → after-hours form/SMS → English staff summary.

Win: permission to show a one-page bilingual conversion demo.

Call 04 · Aventura's Finest · Block 16

Vertical Quote Form

Human call: use their published 24-hour response time as the wedge.

Builder task: detail/tint request → vehicle/service questions → quote packet → confirmation text.

Win: meeting on same-minute booking versus 24-hour response.

Call 05 · Laundromart · Block 07

SMS Reactivation

Human call: ask how repeat customers are contacted across 11 locations and two languages.

Builder task: fake stale customer list → bilingual SMS → pickup/dropoff reply routing.

Win: one-store SMS pilot conversation.

Adversarial rules for tomorrow
  1. No new prospect rows in this section. Add or edit prospects only in prospects-html.js so the source of truth stays clean.
  2. No unverified claims. If the opener depends on a public fact, link it in the call brief before dialing.
  3. No live client data in demos. Fake data until a paid pilot and explicit sandbox boundary exist.
  4. No custom one-off before a block. If the demo cannot map to a block, park it until Evan approves the new block.
  5. No vague next steps. Every call ends as booked discovery, paid pilot ask, followup with date, or logged dead reason.
§ 10 The loop: lead → prototype → sale → repeat.

The internship is not a list of random builds. It is a repeatable factory: find a lead, identify a painful wedge, build a real sandbox prototype, use it to sell, then extract the reusable pattern into the product arsenal.

The operating loop

StageWhat you doExit criteria
1. CaptureAdd every new idea, company, referral, complaint, or weird workflow to the lead queue.Named company, public URL, suspected buyer, observable wedge.
2. ScoreRank it by pain, speed-to-demo, buyer reachability, reuse potential, and API/sandbox readiness.Build now, research more, park, or kill.
3. BriefCreate a one-page call brief with evidence, opener, questions, demo promise, and pilot hypothesis.Ready to call without sounding generic.
4. PrototypeBuild a real sandbox demo using fake data and capped APIs. It must run on a URL.One 60-second wow path works repeatedly.
5. SellCall, show the demo, ask for discovery or paid pilot. Log exact words and objections.Booked, proposed, followup, dead, or new buyer identified.
6. ExtractIf the pain repeats, turn the build into a reusable product block or improve an existing block.New checklist, prompt, route, pricing, or block update merged.

Two-agent operating system

The goal is fewer manual handoffs, not fewer sales calls. Run the agents every morning, then use the output to make sharper human calls.

OwnerInputOutput
Lead AgentNiche, company URL, local list, referral, or raw idea.Ranked prospects, public evidence, wedge hypothesis, call opener, qualification questions, next 5 calls.
Builder AgentBest lead brief, chosen product block, buyer pain, fake data constraints.Cloudflare wow MVP plan, routes to build, prompts, fake data, demo script, deployed URL.
Human SalesLead brief, demo URL, wedge script, objection map.Discovery booked, paid pilot proposed, closed deal, or dead/learned reason logged.
SystemCall notes and demo learnings.Updated product block, stronger prompt, reusable checklist, better next lead score.

Morning agent run prompt

Paste into Claude Code or Codex at the start of the day
Use https://cafecito-ai.com/new-hire as the operating manual.

Enter plan mode first.

Today, run the two-agent loop:
1. Lead Agent: find or refresh 10 leads, score them, pick the top 5 human calls, and write call briefs.
2. Builder Agent: choose the highest-leverage lead and product block, then build or improve one fake-data wow MVP on Cloudflare.
3. Human Sales Pack: output the exact call opener, wedge questions, 60-second demo path, objection answers, pilot ask, and followup message.

Constraints:
- Cloudflare Workers + Hono first.
- OpenAI API + Claude API as model providers.
- Wrangler secrets only.
- Fake data only.
- No production credentials, PHI, payment data, or uncontrolled SMS/voice spend.

Done means:
- top 5 call list
- one updated lead brief
- one demo URL or shipped demo improvement
- one human call script
- one logged next commercial action

Four daily loops

These are not weekday buckets. Run all four every day. The weight changes based on where the best prospect sits, but the loop never pauses.

LoopDaily actionOutput
Lead LoopFind, refresh, or qualify 2-3 leads from hot list, referrals, calls, local search, chamber lists, or client adjacencies.Lead queue stays alive and ranked.
Wedge LoopValidate one painful wedge using public evidence and buyer language from calls/followups.One sharper call brief or kill decision.
Prototype LoopAdvance one sandbox demo toward a 60-second wow path using fake data and capped APIs.Real URL gets more demo-ready.
Sales LoopMove one prospect to discovery, demo, paid pilot, proposal, or closed.Commercial next step scheduled or logged dead.

When a client closes

After close, delivery becomes the main loop. Build the client project out with the same discipline: daily shipped milestone, daily buyer-visible progress, and daily extraction into reusable assets. Keep a smaller lead loop alive so the pipeline does not go cold.

Lead intake template

Copy into the hub for every new idea
Lead / idea:
Source:
Company URL:
Buyer hypothesis:
Observable wedge:
Why now:
Matched product block:
Can we demo this with fake data? yes/no
APIs needed:
Evan sandbox work needed:
48-hour prototype idea:
Call opener:
Kill criteria:

Scoring rubric

ScoreQuestionPass signal
PainDoes this touch revenue, response time, staff overload, missed calls, or customer leakage?Buyer can describe the cost in their own words.
Demo speedCan we show the before/after with fake data in 48 hours?One URL, one scenario, one output.
Buyer accessCan we reach an owner/operator by phone or warm intro?Named person, number, or referral path.
ReuseCan this become a block we sell to the next company?At least three similar businesses would care.
Sandbox readinessCan Evan provision safe APIs and caps without production data?Secrets, fake data, route, reset path, logging.

Prototype acceptance criteria

  1. Real URL: deployed route, not a screenshot or local-only demo.
  2. Fake data: no PHI, payment info, private customer lists, or client production credentials.
  3. One wow path: the demo shows a before/after workflow in under 60 seconds.
  4. API guardrails: capped model, SMS, voice, email, and payment-test usage.
  5. Resettable: Evan or the intern can reset the demo state before every call.
  6. Close-ready: the last screen naturally supports a paid pilot ask.

Rinse-and-repeat weekly target

5new leads
10outbound touches
1live demo
1pilot ask

Kill rules

  • Kill if the pain is not observable and nobody confirms it on a call.
  • Kill if the demo requires production integration before it can be impressive.
  • Kill if the buyer is unreachable after two call blocks and no warm path exists.
  • Kill if it cannot become a reusable block and the one-off value is below $5k.
  • Kill if compliance risk is unclear and Evan has not approved the sandbox boundary.
§ 11 How you know you're black belt.

Concrete artifacts, not vibes. Each belt is a shipped thing, not a feeling. Cross them off in your hub.

WhiteHello-world Worker live at a real .workers.dev URL. Day 1 setup complete.
YellowShipped one mini-product solo to a real URL with Stripe checkout. v0.1 scope held — no scope creep.
OrangeRan a 20-minute discovery call solo. Sent the followup. Logged the outcome with the prospect's exact words.
GreenClosed first paid pilot ($1k+). Wrote the proposal yourself. Defended scope vs. price live.
BlueShipped a client feature unsupervised, end-to-end. Read 1k+ lines of someone else's worker code and edited it.
PurpleBuilt and deployed an MCP server used in a production workflow. Wrote a SKILL.md from scratch.
BrownLed a Teach-Up that taught the mentor something they didn't know. Memo lives in the operator manual.
BlackTwo closed deals, OR $2k MRR running, OR a $10k+ deal you sourced + scoped + built + shipped. Can onboard the next cohort unsupervised. Has identified one new product block worth adding to the arsenal.

If you're at green on building but white on calling, that's normal at Week 3 — push the calling. If you're at orange on both at Week 4, you're behind on the headline goal — shift the next session to the calling motion, not the next mini-product.

§ 12 Make this page better.

This page is v2.0. Every cohort makes it sharper. The first hire (you) becomes contributor #1.

How to improve it

  1. Find the page source at /home/eratner/cafecito-ai/new-hire/index-html.js.
  2. Open a PR against cafecito-ai, or DM Evan R. with the proposed copy.
  3. Fold in: new product blocks (you extracted one), new prospects (you found a great one), updated curriculum (a Greg episode that should be required watching).

What goes here over time

  • More product blocks (Block 11+) as we extract patterns from new client work.
  • More prospect dossiers as we research new verticals.
  • Updated curriculum links as new must-watch content drops.
  • A "results" section: anonymized stats on what cohorts shipped + closed.

Hall of contributors

  • v1 — Evan Ratner (founder) — drafted the doctrine, the curriculum, the §08 blocks, the §09 hot-list methodology, and the first 21 dossiers.
  • Cohort #1 — Evan Seder — May 12 → June 22, 2026. Contributor #1 by virtue of being the first cohort. Will sharpen wedge calls, add new prospects, and ship Block 11.
  • Future cohorts — your name here.

Right now, before you close this tab

  1. Bookmark this page.
  2. Open §02 and create the eight accounts. ~30 minutes.
  3. Watch the first Greg Isenberg episode (link in §04). 25 minutes.
  4. Show up Monday with your gap-finder Worker concept ready to ship.

Build something you'll want to show people on June 22. Vamos.

Stuck? Need help?

Mentor (Cafecito): Evan Ratner · evan@cafecito-ai.com · Tuesdays + flex

Sales co-mentor: Baruki Cohen · Wednesdays + flex

Personal hub: cafecito-ai.com/evan/ (password-gated)

Escalation: 30 min blocked on something a search would have solved? Ask. 2 hours blocked on something nuanced? Definitely ask. Cost of asking is low; cost of spinning is high.