Hey Makers
The results are in — and we have a winner! We’re thrilled to announce @Marco_Handke as the champion of our Designing Intelligent Workflows with Make AI Agents Challenge! His entry didn’t just meet the brief — it raised the bar for what an AI Agent architecture can look like in the real world. Let’s dig in!
The Problem: Too Much Living Rent-Free in Your Head
When you run a business alone, you are the backoffice. Every email thread, every follow-up, every deal update — it all lives in your head, and one missed signal can cost you a client.
Marco knew that rigid if-then automations weren’t going to solve this. The problem wasn’t a lack of tools; it was a lack of judgment. He needed something that could think, prioritise, and act — not just trigger.
The Solution: Meet Mary
Mary is Marco’s AI Chief of Staff, built entirely on Make AI Agents. She’s not one agent — she’s an entire team. Mary leads a family of specialised sub-agents covering calendar, mail, tasks, projects, contacts & deals, Make monitoring, and memory. She runs on Claude Opus 4.6 for high-level reasoning, delegates to Sonnet 4.6-powered sub-agents for execution, and knows exactly when to act on her own versus when to loop Marco in.
Two features make her stand out from the crowd:
- A security gate that performs regex checks on both input and output, guarding against prompt injection at every layer
- Proactive scheduling — Mary runs on a timer, works through her own task list, and surfaces what matters before you even think to ask
Getting the multi-layer orchestration to run reliably was the hardest part: agents calling sub-agents calling tools, all needing to behave consistently in a live production environment. Marco cracked it through rigorous testing and a clear philosophy — handle the trivial autonomously, escalate only the decisions that truly need a human.
What our Judges Loved
The judges were our well-respected and knowledgeable Community Champions. Here’s what they loved about the solution:
A perfect 5/5 across the board — here’s why:
“I love it when people use Make to improve their OWN life. This is a real architecture approach.”
“Looks like he knows what he’s doing, and the solution can be easily expanded. Really nice.”
“There’s even a check for prompt injection attempts. Multilayered, so the agent doesn’t get bloated. Tools, knowledge, and memory — all there.”
Real Numbers, Real Impact
Mary isn’t a proof of concept — she’s been in daily production from day one:
- 2,700+ operations in the past 30 days alone
- Mail, tasks, projects, and memory agents firing hundreds of times each — autonomously
- Marco spends less time on coordination and misses far fewer follow-ups
The most compelling proof? One of Marco’s own clients saw Mary in action and is already planning their own implementation.
The roadmap from here includes a multi-tenant rollout for his core customer group — small insurance brokerages facing the same chaos — and, if tests hold, expanded write-access so Mary can do even more without being asked.
