Elvison
What you're getting
- 18-month timeline broken into 6 phases, each with a hard deliverable
- Stack decision tree for months 1-3 (Claude Code vs API vs no-code)
- Income model per phase (what you charge, to whom, for what)
- The exact 3-agent sequence that generates your first $1k
- Checkpoint test for each phase so you know if you're on track or behind
- One file you fill in to make this your roadmap, not mine
Quick start (5 minutes)
1. Copy the roadmap template block below into a new file: roadmap.md
2. Fill in your current month next to "START:"
3. Mark your current phase honestly
4. Pin the file. Check it every Monday.
# Create it
touch ~/roadmap.md
open ~/roadmap.md
Expected output: blank file. That is the point. You are about to fill it.
The actual thing(s)
Phase 0: Proof of concept (months 0-2)
One agent. One real use case. One person who would pay for it.
# Phase 0 deliverable
- [ ] One Claude agent running locally via claude -p
- [ ] Solves a task you currently do manually (takes you 30+ min/week)
- [ ] You have shown it to 3 people who are not friends
- [ ] At least 1 said "I would pay for that"
Stack: Claude Code subscription only. No API costs yet.
Income: $0. That is correct.
The gotcha here: most people skip Phase 0 and build something impressive instead of something useful. Impressive does not pay. Useful does. You are looking for a boring win, not a demo reel.
Phase 1: First paying user (months 2-4)
Charge one person $99-299/month for the agent output, not the agent itself. They do not run it. You run it. You deliver results.
# Phase 1 deliverable
- [ ] One retainer client. Any amount above $99/month counts.
- [ ] Weekly or daily output delivered via email, Notion, or Slack
- [ ] You have not built a product yet. You are running a service.
- [ ] Agent runtime is < 10 min/day to operate
Charge for: research reports, content plans, competitor audits, inbox triage
Stack: Claude Code + one output channel (email or Notion)
Income target: $99-299 MRR
Running it as a service first tells you what the agent actually needs to do vs what you imagined. Skip this and you will build the wrong product for months 5-10.
Phase 2: Productize the delivery (months 4-7)
Wrap the agent in a UI or scheduled job so it runs without you touching it.
# Phase 2 deliverable
- [ ] Agent runs on a cron or webhook, not manually
- [ ] Output is delivered automatically (email, Slack, dashboard)
- [ ] You can go 48 hours without touching it and it still works
- [ ] 2-3 clients on the same agent (not custom builds per client)
Stack: Node.js cron + claude -p subprocess + one delivery method
Cost to run: under $50/month at 3 clients
Income target: $500-900 MRR (3 clients x $200-300)
This is where most "vibe founders" stall. The agent works in demos. It breaks in production at 2am when nobody is watching. Build the error handling before you scale clients.
Phase 3: First real product (months 7-11)
Stop selling time. Start selling access.
# Phase 3 deliverable
- [ ] A URL someone can log into
- [ ] They can run the agent themselves, no you involved
- [ ] Pricing is per seat or per usage, not per month of your time
- [ ] You have 5+ active users (not signups, active)
Stack: Express or Next.js + Supabase + Claude API (not just Code subscription)
New cost: API spend starts here. Budget $0.03-0.15 per run.
Income target: $1,500-3,000 MRR
The Claude API transition is a real decision. Sonnet 3.7 at $3/million input tokens is the default. Haiku at $0.25/million for anything that does not need reasoning. Do not run Opus on bulk jobs.
Phase 4: One distribution channel that works (months 11-15)
You do not need content marketing, SEO, cold email, and ads. You need one channel that consistently brings in 2-4 new trials per month.
# Phase 4 deliverable
- [ ] One channel only. Pick: IG, LinkedIn, X, or cold outbound.
- [ ] 8+ new trial signups per month from that channel alone
- [ ] CAC (cost to acquire one customer) is documented
- [ ] Churn < 15% monthly
What Elvison uses: IG carousels + keyword DM automations
Time budget for distribution: 3-4 hours/week max
Income target: $3,000-6,000 MRR
If you are not tracking CAC by month 12 you are guessing. The math is: what did you spend in time and money to get one paying user. If that number is higher than your first-month revenue, fix it before growing.
Phase 5: Remove yourself (months 15-18)
The roadmap ends here for a reason. At 18 months the goal is: the product runs, clients are paying, distribution is working, and you are working on it 10-15 hours per week by choice, not 60 hours by necessity.
# Phase 5 deliverable
- [ ] Support handled by docs + async (not you live)
- [ ] Onboarding is self-serve
- [ ] One other person (contractor, co-founder, or agent) handles at least one function
- [ ] You could take 2 weeks off and MRR would not drop
Income target: $5,000-12,000 MRR
Stack: Whatever you built in Phase 3, hardened
This is where the "part-time job" framing from the post matters. At 18 months, most people with a part-time job have gotten two pay raises and are still trading time for money. At 18 months on this roadmap, you have an asset.
Real example
Here is the actual Phase 0 to Phase 1 transition Elvison ran for the Social OS product.
Phase 0 agent (ran it manually for 6 weeks):
echo "Analyze this Instagram competitor: @barn_gym_ie.
List their top 5 content formats, average engagement rate,
and 3 content gaps we could fill." | claude -p --dangerously-skip-permissions
Output: 400-600 word competitor brief. Took 45 seconds. Was doing it manually in 45 minutes.
Phase 1: Charged Barn Gym (Guy Kennedy) $250/month to get a weekly brief like that for 3 competitors every Monday morning. He did not care how it was made. He cared that it showed up.
Phase 2: Wrapped it in a cron job in server.js, auto-emailed every Monday 7am. Now it runs for 6 clients on the same code.
Troubleshooting
"I do not know what agent to build in Phase 0."
Pick one task you do manually that involves reading text and writing text. Research, summarization, drafting, categorization. Any of those works. Do not pick image generation or code generation for Phase 0. Too complex, too much debugging time.
"I have been in Phase 1 for 4 months and cannot get to Phase 2."
You have a client problem, not a tech problem. Your Phase 1 client is probably getting too much custom attention. The fix: document exactly what you do for them each week, then write that as a prompt. If you cannot write it as a prompt, you are doing work the agent cannot do yet. That is fine. Narrow the scope.
"My agent breaks when I am not watching it."
Add this to every agent run:
try {
const result = await runAgent(prompt);
await saveResult(result);
} catch (err) {
await notifySlack(`Agent failed: ${err.message}`);
// do NOT silently swallow errors
}
Slack webhook for errors costs nothing. Silent failures cost clients.
"I am at $2k MRR but stuck."
Check churn first. If you are acquiring 3 clients/month and losing 2, you have a leaky bucket, not a growth problem. Fix retention before fixing acquisition. Pull your last 5 churned users and find the pattern.
"When do I switch from Claude Code subscription to the API?"
When you have 3+ clients who need to run the agent themselves (not just you running it for them). Until then, Claude Code subscription is cheaper and good enough.
Step it up
1. Stack the phases. Phases 0-2 can overlap. While Phase 1 clients pay you for manual delivery, start automating Phase 2 in parallel. You have real use cases to test against.
2. Use the checkpoint test weekly. Every Monday, look at your current phase deliverable. Are all boxes checked. If not, that is your only job this week. No new features until the phase deliverable is done.
3. Build one lead magnet per phase. When you finish Phase 1, write a one-page doc about what you learned. Post it. It becomes your best distribution for Phase 4. The audience trusts the person who shows the work, not the person who shows the result.
What's next
Reply to this DM with STACK and I will send the exact tools I use at each phase, with pricing.
Or hit the newsletter link in bio for the weekly breakdown of what is actually working.
-- Roelof @ Elvison