Why "Vibe Compliance" Will Never Be a Thing
Vibe Compliance is Not a Strategy.
Vibe Coding is the practice of asking an LLM to write software, not really reading it, and shipping it anyway. Fun for weekend projects. Mass casualty event for audit.
Vibe Compliance is what happens when you apply that same energy to professional services. AI generates audit-ready evidence, workpapers, or conclusions — and nobody builds in structured verification, human judgment, or an unbreakable chain of accountability. That's not a bold product thesis. In audit terms, we'd just call that a deficiency.
If you want to see what that looks like in practice, Delve has been providing a pretty vivid case study.
I've been thinking about this problem for a while.
Before ChatGPT was a household name, I was at AuditBoard helping build out their integration platform from the ground up. Before that, I was at Workiva building audit and financial reporting infrastructure where the stakes around data integrity are about as high as they get.
I've spent a lot of time thinking about what happens when you get this wrong.. and who signs the opinion when you do.
That context is why, when we started building Tellen, we made one very deliberate call: we are an AI agent platform for audit firms. Not a "GRC platform." Not a generic "AI for finance" tool. A platform built specifically for accounting firms doing audit, with all the rigor, traceability, and professional standards that entails.
What that actually means in practice.
AI models (and there are a lot of great ones) are genuinely remarkable at extracting meaning from messy client-provided documents, surfacing patterns across control testing populations, and identifying gaps in evidence packages. We use them. We think carefully about which ones are right for which tasks.
But the model is never the point. The point is what you build around it.
Structured data agents that don't produce freeform summaries. They populate typed schemas tied directly to your engagement framework. Every field has provenance. Every extraction traces back to source evidence. Every fact builds up to one cohesive knowledge graph tied together through hundreds of workpapers, templates, and evidence files.
A workflow engine that models the actual lifecycle of an audit engagement: who reviewed what, when, under what criteria, and what conclusion they reached. Not a chat thread. An actual workflow, with states, transitions, sign-offs, and an audit trail that stands up to PCAOB and AICPA scrutiny.
A data model built on the assumption that a conclusion isn't a conclusion until a licensed professional qualified to make that judgment says it is. AI accelerates the work. It doesn't replace the judgment, and it never will.
This is the thing Vibe Compliance gets exactly backwards. The value of AI in external audit isn't "AI does the audit." AI handles the tedious work, clearly, traceably, and with the auditor in the loop at every step. Nothing is left to vibes.
The human-AI interaction problem nobody is solving.
There's a pattern playing out across the agentic AI space right now: teams ship full-blown automation because they can, not because anyone asked for it. The demo looks incredible. The agent extracts data, populates workpapers, draws conclusions, and packages everything up with a bow on top. Fully autonomous. Zero friction.
And nobody trusts it.
Not because the technology isn't impressive — it is. But because jumping straight to end-to-end automation skips the part where a professional actually understands what happened, why the agent reached that conclusion, and whether they'd stake their license on it. In external audit, that's not a nice-to-have. That's the entire point.
We understood this from the beginning. The user experience of how humans work with AI — not how AI works instead of humans — is the design problem that actually matters. Where does the agent surface its work for review? How does a senior interrogate an extraction before signing off? When does the system pause and say "this needs a human decision" versus "this is ready for your review"?
Get that interaction model wrong and you end up exactly where Delve landed: vibed compliance reports papered over real gaps, and nobody caught it until the breach. Sorry Delve — we genuinely like what you're building, but you're making it too easy for us over here.
Get it right and something interesting happens: auditors actually trust the tool. They lean into it instead of routing around it. The data in the system reflects reality because people believe in the process that produced it. And that trust compounds — across engagements, across teams, across the firm.
That's what we've been building. Not because it's a nice-to-have, but because in audit, trust in the process is the product.
One more thing.
We do have a Claude plugin.
For those moments when you just want to ask a question, think out loud, or yes, vibe a little. We get it. Sometimes you just need a great AI assistant in your corner.
The difference is knowing when you're vibing and when you're producing a work product someone is going to rely on. That line matters in external audit. We help firms walk it every day.
The bottom line.
The firms and platforms that get this right: structured, traceable, human-in-the-loop by design, are going to win. The ones that shipped Vibe Compliance and hoped nobody would notice are already finding out what happens next.
External audit is the hardest workflow in professional services. Solve it right, and the rest of accounting is within reach. So watch out. We're coming.
Paul Gerlich is the co-founder and CTO of Tellen. He previously helped build the AuditBoard platform and Workiva's audit and financial reporting infrastructure. He holds a Master's in Data Science (AI/ML) from the University of Illinois Urbana-Champaign.