The agent template I built in 12 minutes that's been running for 3 months
· Avery NXR
We have an unusual workflow at Avery Software that none of the off-the-shelf 7 agent templates exactly matched. So I built a custom one. Total time from "open the agent builder" to "this is running in production": 12 minutes.
That was three months ago. The agent has been running continuously since then. Here's what it does, how I built it that fast, and what changed about the underlying workflow.
The workflow
Every time someone signs up for our private beta waitlist, we want to do three things:
- Send them a personalized welcome email (not a templated mass email — something that references the form fields they filled in)
- Tag them in our CRM with their use case and team size
- Notify our team in Slack with a structured summary so we know to follow up
The off-the-shelf path to this is wiring up Zapier with three actions, calling OpenAI for the personalized email composition, and managing the connections. Estimated cost at our volume: about $30/month for Zapier + $20-40/month for the OpenAI calls. Time to build: probably 90 minutes including testing.
I wanted to see if I could do it in Avery NXR using the agent builder.
The 12-minute build
Minute 0-2: Open Avery NXR → Agents tab → + New Agent → Blank. I picked blank instead of template because I knew the workflow didn't match any of the 7 directly.
Minute 2-3: Drop in a webhook trigger node. Configured it to listen on https://api.avery-internal.com/waitlist-signup (placeholder for the actual webhook). The trigger captures the form payload.
Minute 3-5: Drop in an LLM node. Wrote a prompt that takes the form fields and composes a personalized welcome email. The prompt mentioned the company name, the role, the use case the person described, and what they should expect next. Tested it once with a fake payload. Output looked good.
Minute 5-7: Drop in a Send Email node. Connected our Resend account (which was already in Avery's connector store from a previous setup). Pointed it at the LLM output as the body, the form's email field as the recipient, and a hard-coded "Welcome to Avery NXR" subject.
Minute 7-9: Drop in our CRM connector. Configured an "upsert contact" action. Mapped the form fields to CRM fields. Added a custom tag based on the use case the person mentioned.
Minute 9-10: Drop in a Slack node. Connected to our team's #signups channel. Composed a message template using the form fields plus the LLM-classified use case.
Minute 10-12: Wired it all together in sequence. Tested with a fake payload. Got the email, the CRM update, and the Slack message. Activated.
That's the whole build. 12 minutes from start to running.
Why this took 12 minutes and not 90
Three things compressed the build time:
The connectors were already there. Resend, our CRM, and Slack are all in Avery's 63-connector library. I'd already done the OAuth flow for Slack on a previous agent. The connections were reusable.
The local LLM removed the cost-management overhead. I didn't have to think about token usage, prompt optimization for cost, or rate limits. Run the LLM call. Move on.
The visual builder doesn't require thinking about plumbing. Drag, drop, connect. The orchestration logic is visible. No code editor, no JSON config files, no managing the runtime environment.
This isn't because Avery is magic — it's because the cost of the second-and-beyond agent collapses once you've done one. The first agent you build in any tool takes longer because you're learning the tool. By the fifth, you're at "minutes per workflow."
What's happened in three months
The waitlist signup agent has handled every signup since. Several hundred at this point.
The email response rate is materially higher than templated mass emails would be. Most people who sign up reply to the welcome email with a question or a clarification — which is exactly the engagement we want, because it kicks off the actual conversation.
The CRM tagging has worked perfectly. We know which use cases are coming in, which team sizes, which industries. The structured data lets us prioritize follow-ups intelligently.
The Slack signal-to-noise has been right. Signups show up in real time. Our team responds to interesting ones within hours of the form submission. We've closed multiple early customers because someone on our team saw the Slack ping and replied within the same business day.
The agent has needed maintenance exactly twice in three months: once when we added a new form field that needed mapping, and once when I updated the welcome email prompt to mention a new feature. Both edits took under 2 minutes.
What this generalizes to
The thing I keep coming back to is the difference between "AI is something I subscribe to" and "AI is something I build."
When AI is something you subscribe to, every workflow is the vendor's workflow. You configure within their abstractions. You're stuck with what they thought to build.
When AI is something you build — and the cost of building drops to 12 minutes per custom workflow — you start asking different questions. Not "is there a tool for this?" but "should we have an agent for this?" The answer is usually yes for anything that repeats.
Avery NXR's combination of visual builder, local model (no cost-per-execution overhead), and pre-built connectors makes the build cost low enough that the calculus shifts. You're not gated by what's worth paying for. You're gated by what's worth your attention.
What I'd suggest
If you've been using AI tools and feeling the limitation of "the vendor decides what I can do," install Avery NXR and try building one custom agent.
Pick a workflow you wish was automated and isn't. Don't pick something ambitious — pick something small. The signup-flow agent above is a great pattern: one trigger, three actions, done.
Build it. Watch it run. See whether the 12-minute experience changes how you think about what's worth automating.