I’ve been having a lot of conversations lately about AI governance. Finance leaders, internal audit teams, IT directors — all trying to figure out how to put guardrails around tools that are moving way faster than their policies can keep up with. What I keep noticing is that most of these organizations are still at the very beginning of the curve. They’re using AI like a fancy search engine.
Then a developer named Bassim Eledath published a blog post that put a framework to something I’d been seeing. He calls it the 8 levels of agentic engineering.
The core idea is simple. AI capability is no longer the constraint. The bottleneck is knowing how to use it.
Anthropic’s team shipped a product called Cowork in 10 days using the same AI models that other teams can barely get past a prototype with. The difference wasn’t the model. It was the methodology. That’s the whole point. And it matters a lot for finance.
The Automation History Every Accountant Knows
We’ve been through this before. Spreadsheets in the ’90s. ERPs in the 2000s. RPA in the 2010s. Each wave shifted what accountants do — not whether they’re needed.
The RPA wave is worth thinking about now. The promise was: describe what you want, hook it up, and it just happens. But RPA required you to specify every single rule. It was brittle. When something changed it broke.
I’ve been seeing posts from people who lived through the RPA era saying the same thing: agentic AI is what RPA was supposed to be. You give it a goal. It figures out the steps. It uses tools. It self-corrects. That’s a fundamentally different ceiling.
The 8 Levels — A Quick Map
Here’s the framework translated for a finance audience.
Levels 1 and 2 are where most finance professionals are today. Tab completion and AI chat. You’re using ChatGPT to draft emails or summarize documents. It’s useful. But you’re leaving most of the value on the table.
Level 3 is context engineering. The quality of what you get out of AI depends almost entirely on what you put in. A vague prompt gets a vague answer. A well-structured prompt with the right context and examples gets something you can actually use. I’ve covered this in earlier lessons and it’s one of the highest-leverage skills you can build right now.
Level 4 is compounding engineering. Every time the AI makes a mistake you capture the lesson and update your instructions so it doesn’t repeat it. Think of it like updating your SOPs after a close issue — except the SOP actually gets read. Your AI setup gets measurably better every session.
Level 5 is MCP and skills. This is where AI stops just advising and starts acting. You connect it to your ERP, your bank feeds, your reporting tools. At EverydayCPE I use a skills library that helps me produce these lessons. Right now I have Claude working on a Power BI report — connected directly through MCP. It’s not describing what to do. It’s doing it. That’s what Level 5 feels like.
Levels 6, 7, and 8 are the frontier. Automated feedback loops. Background agents running while you sleep. Eventually teams of agents coordinating with each other. Nobody has fully cracked levels 7 and 8 yet. But level 6 is closer than most finance teams realize.
What This Means for Accountants
Two things stand out to me.
First is the close cycle. Month-end close is repetitive rules-driven work with verifiable outcomes. That’s exactly what agents are designed for. Journal entries drafted automatically. Reconciliations run overnight with exceptions surfaced for review. Flux analysis ready before your first meeting. Teams that build agentic workflows into the close are going to compress their timelines in ways that are hard to compete with.
Second is governance. When an AI agent autonomously posts journal entries and updates reconciliations it’s not just a software tool anymore. It’s a control. Most organizations don’t have a framework for evaluating it that way. The 8 levels give you a vocabulary for asking the right questions: What level of autonomy does this agent have? What are its error-correction loops? Who reviews its outputs?
Key Takeaways
- Most finance professionals are at Levels 1–2. The big leverage jump comes at Levels 3–5.
- Context engineering (Level 3) is the highest-ROI skill available to finance teams right now.
- Compounding engineering (Level 4) makes every session better than the last — like an SOP that actually gets followed.
- Connecting AI to your systems (Level 5) is where workflow automation becomes real.
- Agentic AI in financial processes needs to be treated as a control — not just a productivity tool.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply