How We Hit $2K MRR Letting People Deploy AI Agents in 60 Seconds

This is a build-in-public case study of [RapidClaw](https://rapidclaw.dev), a managed AI agent hosting platform we built in late 2025 / early 2026. We hit $2K MRR in our first few months. Here is the real story — what we built, who actually signed up, the channels that converted, what we undercharged on, and the playbook for the next 10×.

If you're building anything in the AI agent space, the lessons here transfer directly. If you're considering AI agents for your own work, the use cases that resonated will tell you which problems are actually paying.

TL;DR — what worked, what didn't

• The product: describe an AI agent → it deploys to Telegram, Discord, or WhatsApp in under 60 seconds. No Docker, no VPS, no API key juggling.
• The actual users: small agency owners, not developers. We mis-targeted at launch.
• What converted: observability dashboard > deploy speed > channel coverage.
• Channels that worked: Discord and WhatsApp support tripled signups vs Telegram-only.
• What we got wrong: undercharged ($19 should have been $49). Built advanced features before basic Gmail/Calendar connectors.
• The next 10×: Gmail/Calendar connectors, shared agents catalog, team seats, raised pricing tier, distribution to small agencies specifically.

What RapidClaw actually does

You describe an AI agent in plain language. It deploys to Telegram, Discord, or WhatsApp in under 60 seconds. The agent runs on managed infra we own; you get a dashboard that shows what it's doing, what it's spending, and what tools it's calling.

No Docker. No VPS. No API key juggling. No "set up your serverless environment first." You sign up, describe the agent, pick a channel, and the agent is live.

We built it because we spent too many hours wrestling with broken local AI agent setups — unreliable tunnels, missing dependencies, configurations that silently stopped working. If we hated that, other people had to hate it too. We were right about the pain. We were wrong about who felt it.

The first 90 days, in detail

A week-by-week look at what actually happened.

Week 1 — "Hello world"

Stand up the basic infra. A barebones agent deploying to Telegram on our own servers. The "hello world" moment was sending /hi in Telegram and getting a reply from an agent we'd defined in a config file 30 seconds earlier.

This was the moment we knew the product had a pulse. The 30-second loop was real and felt magical.

Week 3 — Waitlist live

Landing page on Vercel. Waitlist form. One paragraph of "what is this." Posted in two indie hackers communities and on X.

200 signups in ten days. Almost all from a single Indie Hackers post. We were surprised — we expected developer interest. We mostly got non-developer interest. People asking "can it answer my customer support emails?" and "can it follow up with leads?" Not "can it run my dev workflow?"

Week 5 — First paid beta customer

Stripe live. $19/month tier. Manual onboarding via Telegram chat with us. First customer was a small agency owner who wanted an agent that would triage incoming client emails. They had been doing it manually for three years. They paid us $19 to make it stop.

Week 8 — Ten paid customers

Same shape. Mostly small agencies. A few solo consultants. Two indie founders. One developer (us-shaped persona). The pattern was clear — we'd built for developers, but the people paying were operators.

Week 12 — $2K MRR

A mix of $19 plans and a few $49 "more channels + more agents" plans we'd added in week 9. Margins were tighter than we hoped once token costs were factored in. Net contribution per customer was positive but smaller than the headline number.

What surprised us

Three big surprises shaped the next phase.

1. The users were not developers

We built it for ourselves — developers sick of DevOps for our agents. The people signing up were small agency owners with statements like:

I want an AI agent that handles my client emails. I don't care how it's hosted. I just want it to work.

The pitch that resonated was not "no Docker." It was "it works in 60 seconds, you don't need to think about infra."

This shifted everything. The marketing copy. The onboarding flow. The features we prioritized. The channels we showed up on.

2. Observability mattered more than features

We thought we'd win on speed of deploy. We actually won on showing users what their agents were doing and spending.

Giving a user a dashboard that says "this agent replied to 47 emails this week, classified 12 as needing your attention, and cost you $4.20" — that was the moment trial users converted to paid. Not the deploy speed. The visibility.

The lesson generalized: in AI products, trust is a UX problem. Users don't trust models out of the box. They trust models when they can see what the model did and what it cost.

3. The channel mattered more than the product

Telegram support was fine. Discord support doubled signups. WhatsApp support tripled them — because WhatsApp is where our agency users already lived with their clients.

We had thought of channels as a "supported integrations" line. They were the actual product. The agent code was 80% the same across channels. The distribution was 100% different.

What we got wrong

Three big mistakes worth not repeating.

1. We shipped pricing too late

For three weeks we accepted interest but didn't take money. The reason was internally rational ("we're still beta") and externally a disaster — by the time Stripe was wired up, we'd lost the urgency from half the waitlist. People who were hot in week 1 had moved on to alternatives by week 4.

Lesson: ship pricing with the landing page, not after. Even if the price is "early access $19/month with a money-back guarantee," the act of charging filters serious users from polite interest. See Ship a SaaS in 48 hours for the playbook we should have followed.

2. We built a feature we didn't need

We spent two weeks on custom tool integrations for "advanced" users — letting power users wire arbitrary OpenAPI specs into their agents.

Those users didn't show up. The users who did show up wanted simple Gmail and Google Calendar connectors. We had built the high-end before the low-end. We're now playing catch-up on connectors that should have been week-one features.

Lesson: build for the user who actually pays you, not the user you imagine. The first paying customer's needs are the priority list.

3. We undercharged

$19/month felt right to us. It was wrong.

The agency users we'd attracted were saving 5-10 hours per week of manual triage and drafting work. At even a $50/hr blended cost, that's $1,000-2,000/month of value. Charging $19 captured 1-2% of that value. The agency advisors we talked to all said the right starting price was $49 — and probably $99 with annual discounts.

Raising prices after the fact is friction. Every existing customer either gets grandfathered or feels burned. The right move was to start at $49 and offer a $19 "personal" tier as a loss-leader funnel.

Lesson: if you're building for SMEs solving real time-eating problems, charge what the time is worth, not what the infra costs. We'll be using this lesson on the next pricing change.

What worked — the channels that converted

A breakdown of where the first 50 paying customers came from:

• Indie Hackers post (~30%) — The single highest-conversion post. A "we just shipped this" post in the right community. Lesson: pick one community where you'd genuinely hang out anyway, and build trust over time.
• Personal X / Twitter network (~25%) — Tijo's existing audience from previous projects. Build-in-public posts shipping milestones. Lesson: distribution compounds. Every previous shipped project pays off in the next launch.
• Direct messages to specific people (~20%) — Reaching out to people we knew who had the exact problem. Lower volume, much higher conversion. Lesson: 10 personal asks beat 1,000 impressions.
• Search (~15%) — A few posts ranked for "AI agent hosting" and "deploy AI agent Telegram." Slow but steady. We're investing more here in the next quarter — see vibe coding for SEO for the broader play.
• Word of mouth (~10%) — Existing customers telling other agency owners. The most flattering segment, the most fragile. Asking customers for referrals at the right moment converted maybe 1 in 5.

We did not run paid ads. They were not the right move at this stage — we hadn't earned the LTV math yet.

What the next six months look like

The roadmap, public:

Gmail and Calendar as first-class

The most-requested feature, the one we should have shipped at launch. Currently underway. Lesson learned: build the boring connectors first.

Shared agents catalog

A directory of working agent templates. Clone a "client email triage" agent, customize the prompt, deploy. Reduces the cold-start problem and gives users a starting point that already works.

Team seats

Currently every customer is a single user. Agencies want to share agents across their team without forwarding API tokens. Multi-seat is the next conversion unlock for our larger customers.

A higher-priced tier

We'll introduce a $99/month "Agency" tier with team seats, more channels, more agents per workspace, and priority support. The current $19 plan stays for personal users.

More channels

WhatsApp Business API native support. Slack apps. Eventually iMessage (when Apple opens that up).

A better dashboard

Spend forecasting. Agent performance trends. "Your agent saved you ~7 hours this week" framing. The dashboard is the trust layer; investing here directly grows MRR.

The honest take

$2K MRR is not venture-scale. It's also not nothing. It's a signal that:

• The problem is real.
• The buyers exist.
• The channel works.
• The product solves enough of the problem to make people pay.

The next milestone is 10× this — $20K MRR. The path is not more features. It's better distribution to the audience we now know resonates. Specifically: small agencies. Specifically: ones drowning in client email triage and proposal drafting. Specifically: ones that already use Telegram, Discord, or WhatsApp with their clients.

We're not chasing developer adoption anymore. We were wrong about that and the data corrected us.

What I'd tell another indie founder building in this space

Five things, distilled:

1. The deploy-speed pitch is a hook, not a moat. Everyone will be 60 seconds eventually. The moat is reliability, observability, and channel coverage.
2. Talk to the first 10 customers individually. Every single one. We learned more from week 5 than from any "user research" session.
3. Charge more than feels comfortable. If your customer is saving 10 hours/week, $99/month is a steal. $19/month feels nice. It is wrong.
4. Pick a niche channel mix. "We're on every channel" is generic. "We're the best at Telegram + Discord + WhatsApp for small agencies" is specific. Specific wins.
5. Ship pricing with the landing page. Real money sorts real demand from polite interest.

FAQ

What's the tech stack behind RapidClaw?

OpenClaw (open source agent framework) running on managed infra. Next.js dashboard on Vercel. Postgres for state. Redis for queues. The agent runtime is in MicroVMs for the higher-tier plans (Builder Sandbox, White-Glove). We have a separate post on the stack on humanai.news.

How much does it cost you to run an agent?

Roughly $0.50-2.00 per active customer per month in token costs, with infra costs amortized across the customer base. Margin is positive but tighter than a pure-software product because of the variable token cost.

How does RapidClaw compare to building it yourself?

Building it yourself costs you DevOps time. If your hourly rate × the time it takes you > our subscription cost, we win. For most non-engineer users (agencies, consultants, indie operators), the math is wildly in our favor.

Is RapidClaw open source?

The agent framework underneath (OpenClaw) is open source. The managed hosting layer, observability dashboard, and channel integrations are proprietary. The model is "open core, managed cloud" — same as many SaaS infra products.

What channels does RapidClaw support?

Telegram, Discord, and WhatsApp at the time of writing. Slack and email are on the roadmap.

Can I migrate off RapidClaw if I want to self-host later?

Yes — the agent definitions are portable. You take your prompt, your tool wiring, and your conversation history with you. We do not lock that in.

Who's the team behind RapidClaw?

Tijo (founder, growth/distribution side, also the author of the humanai.news content brand) and Brandon (co-founder, infra side). Two people. We do not currently have outside investors.

The bottom line

If you're building something in this space — AI agents, indie SaaS, anything where the buyer is an SME with real time-eating problems — the lessons here transfer:

Ship fast. Charge from day one. Talk to the first 10 customers personally. Pick a specific niche channel mix. Charge more than feels comfortable. Build the boring connectors first.

For the broader playbook: What is vibe coding, Ship a SaaS in 48 hours, Why agencies are replacing Zapier with AI agents, and 30 AI agent ideas you can deploy in 60 seconds.

Try RapidClaw at rapidclaw.dev — a personal AI agent live in 60 seconds. For ongoing coverage of the AI tooling space: humanai.news.