A team I worked with spent months planning an AI initiative to automate their existing process. The process was manual, slow, and familiar. Everyone agreed it was a good candidate for automation. The plan was solid. The technical approach was sound. They were doing exactly what they'd been asked to do.
Until someone stepped back and asked a different question: should we be running this process at all?
The answer was no. The team had been reacting to a problem they could have been predicting and preventing. But no one had questioned the premise because the premise had been working well enough for long enough that it felt like a fact.
AI didn't create that blind spot. The team had been operating around it for years. What AI did was make the blind spot worth questioning before further investment.
My colleague Dave Mosher makes the case that if an organization's intent isn't legible to a machine, it probably isn't as legible to its people as anyone thinks. He's right, and his piece on organizational observability is worth reading alongside this one. It raises the next question: once AI makes the misalignment visible, what do you actually do about it?
AI makes your teams more productive, more capable, faster than ever. That speed is real and the throughput gains are genuine. But throughput without alignment is noise at scale.
What AI actually amplifies
Confident misalignment
Here's what I keep seeing. A team picks up a goal, interprets it through their own context, and runs with it. Before AI, they'd build a proposal, circulate a draft, and get feedback at human speed. Misalignment had time to surface in conversation. Now the same team generates a polished strategy deck and a prototype before anyone outside the team has read the brief.
The output looks so credible that it masks the underlying question of whether the team is solving the right problem. Dave Mosher calls this "ruthless sincerity"—an agent will pursue a local maximum with total confidence.
But the same dynamic applies to the humans using these tools. The person presenting an AI-assisted strategy may feel more confident. The audience may receive it as more credible. And the gap between appearance and reality widens.
AI also exposes the cracks in role clarity and process maturity
When anyone can draft a product spec or spin up a prototype, the question of who should gets lost in the excitement of who can. And when you try to automate a process held together by tribal knowledge rather than documentation, the automation surfaces every undocumented assumption, every handoff that only works because one person knows what to do next. But one pattern is less obvious—and worth watching.
The slow erosion of ownership
I haven't seen many teams explicitly say "the AI suggested it" as a way to dodge accountability. That framing sounds dramatic, and the reality is quieter. What I have seen is output thrown over the wall without the kind of critical review it needed. Work generated fast, accepted at face value, passed forward. Not malicious. Not even conscious. Just a gradual blurring of the line between "the tool produced this" and "I own this."
It's early. As agentic workflows scale and AI handles more of the execution, this erosion will accelerate. The risk isn't that someone explicitly deflects blame onto a model. The risk is that ownership becomes ambiguous—nobody decided, the process just happened, and the output looked professional enough that nobody stopped to ask who was actually accountable for it.
AI doesn't absorb responsibility. If your humans weren't making clear decisions before, AI gives them a faster way to not make them.
The forcing function
The act of implementing AI well forces you to do the same work as fixing your organization. AI just gives you a concrete reason to finally do it.
The diagnostic
AI can actually help here, by surfacing dysfunction without judgment.
The clarity test
When you try to articulate goals for an AI workflow, you discover whether your team actually agrees on what "good" looks like. If you can't state the goal clearly enough for a system to act on, it wasn't clear enough for your people either.
The workflow mirror
When you try to automate a process, every ambiguity in handoffs, ownership, and decision criteria becomes a blocking error. A human can muddle through an undocumented process because they can read the room and make a judgment call. An agent can't.
The alignment audit
AI can review existing artifacts—strategy docs, OKRs, meeting notes—and surface contradictions without anyone losing face. " These four documents describe four different priorities" hits differently coming from a tool than from a colleague.
None of that matters if no one acts on it.
The alignment conversation
Surfacing the dysfunction is the easy part. Converting the signal into alignment is the actual hard work, and AI can't do that for you.
I've watched teams stall at this exact point. The diagnostic surfaces a misalignment that everyone can now see, and the room gets quiet. Not because people don't care. Because questioning the way things have always been done is genuinely uncomfortable—even when the evidence is right in front of you.
There's comfort in established paths. You can see them. You know where they lead. Being told "you have the autonomy to assess this from a different angle" sounds like freedom, but it can feel like being pushed into the void.
What I've seen break through that resistance isn't a presentation or a mandate. It's someone walking alongside the team and modeling what good looks like in practice.
Showing them that revisiting a process doesn't mean declaring it wrong—it means asking whether there's been an evolution. Showing them there are established frameworks for navigating open questions: opportunity assessments, hypothesis testing, structured experiments, even just better questions. The world feels wide open, but there are guardrails if you want them.
This is what "humans in the lead" actually looks like. Not approving machine output, but instead making the decisions AI can't make for you.
To do that, you need someone with the authority to convene the conversation—not the person who found the problem, the person who can decide. You need a structured format that converts "here's what's misaligned" into "here's the decision we're making." You need enough psychological safety for people to say "we've been operating on different assumptions" without it becoming blame. And you need the willingness to actually choose, which means letting go of the other options.
One team I worked with spent several rounds refining an AI strategy—operating models, change management plans, adoption metrics—all built on an assumption no one had tested: that the business case was already settled and they were just optimizing for it.
It took human-to-human conversation, not AI output, to surface the real problem. The initiative didn't yet have a business case. Once we aligned on the need for business justification, the work was rescoped around the appropriate measurable outcomes and quickly moved to pilot. The most valuable output wasn't the strategy. It was the organizational clarity the process forced.
Five questions to ask before you scale AI
Before you invest in more AI throughput, start with these.
- Can you describe the goal of your most important initiative in one sentence that every person on the team would agree with? If not, AI will amplify ten different interpretations at once.
- If you automated your current workflow tomorrow, whose decisions would the AI be making? If you can't name the person, that's the accountability gap that AI will widen.
- Do your leadership team's priorities survive contact with each other? Put the top priorities from each executive on a whiteboard. How many conflict? That's what AI will amplify.
- Is your documentation a foundation or a fiction? If you tried to train an AI on your team's processes, SOPs, and strategy docs today, would it learn how your org actually works, or how it pretends to work?
- When the AI surfaces a contradiction, who has the authority and the willingness to resolve it? If the answer is "nobody" or "it depends," that's the real blocker. Not the technology.
If these questions are hard to answer, that's not a reason to slow down on AI. It's a reason to start with the alignment work that makes AI adoption actually pay off. The technology is ready. These five questions tell you where to begin.
Jen Tedrow is a Director of Product Management at Test Double, and has experience in product strategy, AI adoption, organizational change, and executive alignment.
