Why Stuffing AI Into Workflows Has a Ceiling
By Daniel Koo - Head Of Product
05/20/2026
The most popular form of AI adoption right now is also the most cautious: take an existing workflow, find the slowest step, drop an AI tool into that step. Output gets faster. Everyone is happy.
The shape of the work is unchanged. That's the ceiling, and it's the part nobody is talking about yet.
The pattern is the same in every function
Walk through almost any function in a mid-sized organization and you'll see the same shape.
Marketing has a content team that drafts copy. Someone added an AI drafting tool. The drafter now starts with an AI draft and edits it, instead of writing from scratch. The approval chain (one editor, one stakeholder, one final sign-off) is the same chain it was before.
Finance has a reconciliation team. Someone added an AI matching tool. The team now reviews AI-flagged matches faster than they could match line by line themselves. The exception process (escalate to manager, manager reviews, manager decides) is the same process it was before.
Customer service has a triage team. Someone added an AI summarization tool. The team now reads AI-generated summaries before opening tickets, and the average ticket-to-resolution time drops. The escalation tree (tier 1 to tier 2 to tier 3) is the same tree it was before.
Engineering has a code review process. Someone added an AI code completion tool. The team writes routine code faster. The review process (peer review, then QA, then deploy) is the same process it was before.
In every case, one step got dramatically faster. In none of those cases did the rest of the work change.
The productivity gains at the step level are real
The productivity gains are not exaggerated and they are not imagined. When a step that used to take a person two hours now takes them twenty minutes, that's a 6x improvement at that step. Multiply across a team, across a quarter, and the time savings are substantial.
This is what most leaders mean when they say AI is delivering. They're pointing at the step. The step got faster. They're correct.
The workflow absorbs the gain
The catch is that workflows are not a sum of steps. They are a structure of steps with handoffs, approvals, escalation paths, and decision rights distributed across roles.
When you make one step faster, the rest of the structure absorbs the gain. The team draws breath at that step. They have more time for the next step. They process more volume through the unchanged pipeline.
The pipeline itself doesn't change shape. The handoffs are the same. The approvers are the same. The escalation paths are the same. The decisions still route through the same humans, in the same sequence, with the same boundaries.
Output gets faster. The work doesn't change shape.
The ceiling is in the workflow shape, not the AI tool
This is the part that's easy to miss.
Most workflows in use today were designed when humans were the only available option for every step. That assumption is built into the structure. Approvals are sequential because attention is human and serial. Escalation paths are hierarchical because authority lives in roles, and a role is a person. Handoffs are explicit because tacit knowledge requires conversation.
When AI gets dropped into one step, it inherits those constraints whether they apply to it or not. It produces output sequentially because that's what fits between the prior step and the next one. It waits for human approval because that's where the workflow expects review. It escalates the same way the workflow expects escalation, even when an exception is something AI itself could resolve.
The AI tool is operating inside a structure that wasn't designed for it. The result is real productivity at the step level, capped at whatever the workflow shape can absorb.
That's the ceiling. It's not a ceiling on the AI tool's capability. It's a ceiling on what the surrounding workflow can let the AI tool do.
The ceiling moves but doesn't go away
The first response a lot of organizations have to "tools have a ceiling" is to reach for a different tool. Maybe agents. Maybe a copilot. Maybe a chatbot pointed at a specific use case.
Each of those has a different shape. They share the same assumption: the workflow itself is fixed, and the AI is the variable being added to it.
The ceiling moves. It doesn't go away.
That's worth walking through one shape at a time, which is what we'll do next.
Head Of Product, Arkane Digital
Daniel leads product strategy and system design for AI platforms, focusing on turning complex capabilities into structured, usable systems that support real business operations.