Built a 9-agent playground anyone can try in 30 seconds

I built /play over the weekend. Nine of the agents Automatq sells, each running live on Claude or GPT, each demo-able in 30 seconds on cold sample data. No signup. No card.

This is one of those things you can't talk a prospect into believing. They have to feel it. Description is weak proof — "our agents draft quotes in your voice" sounds like a brochure line every consultancy types out. Watching the Estimator agent read a bathroom-reno quote request and produce a structured estimate with line items, timeline, and a warm sign-off in 8 seconds — that's a different conversation.

The roster

• The Receptionist — already running live on the site, voice mode. Skip in the playground.
• The Estimator — paste a quote request, get back a draft quote in the owner's voice.
• The Follow-Up — watch it write five personalized nudge emails for five quiet leads.
• The Scheduler — natural-language meeting request → three slots + confirm email + day-of SMS.
• The Reporter — paste a rough activity log, get back a Friday digest email.
• The Status Concierge — generate a weekly client-facing project update.
• The Reviewer — three sentiment scenarios (positive, mid, negative), watch it pick platform + timing + write the right message.
• The Voice — pick a topic and format, get a draft post in the brand voice.
• The Onboarder — new client signs → NDA + scope + intake + kickoff invite + welcome email, all generated in one trigger.

What this took

About a week of build. Most of the work was the prompt engineering, not the wiring. Each agent gets a focused system prompt that grounds it in: (1) the brand voice constraints, (2) its specific job, (3) format expectations. Then a single /api/agent endpoint streams whatever agent the visitor selected.

The page itself reuses every existing brand pattern — same gradient cards, same dark glass, same SSE streaming reader I wrote for the Receptionist chat. About 350 lines of new component code; the rest was already there.

Why this is the highest-leverage thing on the site right now

Two reasons.

One: it's an unfair sales asset. When a prospect on a Cal call says "okay but what does the Estimator actually do?" — I can deep-link them mid-call to /play?agent=estimator, watch them paste their actual quote request, and watch the close rate go from 30% to 80%. Nobody else in my market has this.

Two: it's organic distribution. Every agent output is screenshot-able. Every fixture is shareable. "Watch this agent write a contractor follow-up email in 4 seconds" is a Twitter thread that books me three calls. The playground is content-marketing infrastructure disguised as a demo.

What's next

I haven't built the analytics layer yet — I don't know which agent gets clicked most, which gets re-run, which gets shared. That's the next bit. Once I know which agents are actually qualifying prospects, I'll deepen those and trim the ones nobody touches.

Also coming: a tool-use mode where the agents don't just describe the work — they actually do it. The Estimator drops the quote in the prospect's CRM. The Voice publishes the post to LinkedIn. The Onboarder fires the contract via DocuSign. The playground becomes the actual product, free-tier capped, paid-tier metered.

For now: nine agents, free, no signup, live. Go to /play and break them.