AI + Data Engineering Leader | Simplifying Complex Tools | Enabling Smarter Decisions with Custom Assistants & Automation | Bridging Business & Technical Worlds with Scalable, Human-Centered Tools
Many “AI agents” I see today… aren’t agents at all. They’re workflows in disguise. I recently read an article by Jake Jones at Flank called "Stop Calling Workflows ‘Agents’ – A Guide to Real Agentic AI." He nailed something I’ve been talking about at work for months. Too many products (and internal pilots) are really assistants or workflows dressed up with the “agent” label. That might make for a nice demo, but it creates three problems: False confidence → leaders think we’re further ahead than we are. Eroded trust → when “agents” collapse, people conclude “AI doesn’t work.” Missed opportunity → we never invest in the guardrails that make true autonomy possible. Here’s how I now frame it (inspired by Jake, adapted to how we think about AI at work): Level 0 – Automated Workflow: Pre-baked, brittle sequences. Level 1 – AI Assistant: Drafts, summarizes, and classifies. Human drives. Level 2 – Human-in-the-Loop Agent: Can plan/act, but pauses for approval. Level 3 – Human-on-the-Loop Agent: Operates autonomously within guardrails; escalates only on exceptions. And if you’re not measuring things like: Unattended Completion Rate Obstacle Recovery Rate Mean Time to Human Policy Breach Rate …then you’re not talking about Agents. You’re talking about Assistants. And that’s okay — Assistants are valuable! But they’re not autonomous. Language matters. It shapes budgets, expectations, and adoption. If we become sloppy with the term “agent,” we risk undermining the trust necessary to scale AI. 👉 How is your org drawing the line between Assistants and Agents? 👉 Have you ever stopped an “Agent” project once you realized it was really just a software tool? (Link to the article: https://lnkd.in/g4ZBAnPg)
Erin Sanchez
2w
Nick, you’ve got me thinking… I write about AI/tech daily and I think people often use “assistant” and “agent” interchangeably, especially when talking about automation. This is an important distinction I need to keep in mind when receiving a creative brief!
ML+Data Engineer III @ Blue Origin | AWS, Databricks
2wWell written! I'd add that agents are what is being sold left and right. Everyone thinks they need "agents" even if the workflow doesn't make a decision!