Full-Stack AI Engineer (AI-Native)

About AgentCollect

B2B debt collection powered by AI agents

Tech description:

Stack: Python, TypeScript, Postgres, Redis, AWS, Stripe, Twilio, RetellAI, Anthropic Claude.

Our AI agents handle every part of the recovery workflow: outbound voice calls (RetellAI), email sequences, contact discovery, negotiation logic (Claude), and payment processing (Stripe). Each enterprise customer gets isolated agents with their brand voice, dispute playbook, and escalation rules.

Hard problems: voice agents that handle dispute themes specific to each Fortune 500 vertical (Dell warranty disputes, FedEx shipping disputes, Microsoft licensing disputes), regulatory compliance across 50 US states plus GDPR, real-time payment plan negotiation under company policy.

Job description:

We're AgentCollect. We collect overdue B2B invoices with AI agents instead of call centers. Real revenue, real enterprise clients, a small team that ships every single day.



**Why I'm building this**



I started AgentCollect because B2B invoices go unpaid for a dumb reason: the only options were aggressive third-party call centers that torch the customer relationship, or chasing it manually until everyone gives up. We replace that with AI agents that actually recover the money while staying respectful enough that the debtor stays a customer. It works, enterprises pay for it, and there is a lot left to build.



**The stack you'll work in**



RetellAI voice agents, Twilio / Telnyx telephony, Stripe for payments and reconciliation, Laravel + React across the app and the debtor / client portals, PostHog for session analytics, and you direct AI tools (Claude Code, Cursor, agents) every day to ship. Language-agnostic: prior Laravel is a plus, not a gate, we hire people who ramp fast with AI.



**How we work, so you know if it fits**



* We plan first, then iterate in code. We don't paste a prompt and hope.
* AI token cost is irrelevant to us. We'd rather spend tokens and get it right than cut corners.
* We'd much rather hear "I don't know yet, here's how I'd find out" than a confident wrong answer.
* You own what you touch, end to end.



**Your first 90 days**



* **Day 30:** you've shipped real improvements to our voice agents, hardening the retry and human-handoff logic so fewer collection calls drop.
* **Day 60:** you own a system end to end, for example the PostHog session-replay detector that flags broken UX on the debtor and client portals before a user reports it.
* **Day 90:** you own a revenue-facing slice, for example the Stripe billing and reconciliation flow, and you're shipping to enterprise clients weekly.



**Who you are**



* You've shipped real production software (any stack).
* You actually direct AI tools, you don't just autocomplete with them.
* You reason about product, not just code. You ask the sharp question before you build.



**Our application is 5 minutes. Really.**



No 2-hour take-home to get in the door. We want to see how you START a problem:



1. Open a project in plan mode (Claude Code, Cursor, or plan in writing) and record your screen for up to 5 minutes (no face cam needed).
2. Plan how you'd automatically catch a UX bug specific to AgentCollect from our PostHog session replays, before any user reports it. Tell us what you'd need to know that you don't yet.
3. Send us two links: a GitHub repo with your PLAN.md, and your screen recording.



A sharp 5-minute plan beats a polished 2-hour submission that never framed the problem. If your plan is strong, we hand you real data and you build it for real.



Full details and the exact problem are in the README: https://github.com/wwwidr/hiring-challenge



Apply here and we'll send you the 5-minute challenge. Looking forward to seeing how you think.



John, Founder, AgentCollect