Are Project Managers Still Needed in the Age of AI?

I went in half-skeptical, half-curious. I came out with a working automation — and a much clearer conviction that the project management profession is not shrinking. It is evolving into something far more demanding.

The Experiment: Letting AI Take the Wheel

Like many practitioners, I have been hearing the headlines: AI will replace knowledge workers. Automation will make middle management obsolete. Rather than debate it from the sidelines, I decided to run my own hands-on test. I set up a workflow using Activepieces — an open-source automation platform — and wired it together with AI agents designed to surface only the information I actually needed, filtering out the relentless noise of today's information-saturated environment.

The goal was personal productivity: build something that keeps me focused, gives me signal instead of noise, and does it without requiring me to write production-grade code from scratch. Step by step, guided by AI prompts and agent outputs, I got there. The system works. And here is the part worth sitting with: I pushed JSON configurations into the pipeline that I did not fully understand.

That sentence deserves to be read slowly. I — a project manager with over two decades of experience coordinating engineers, architects, and delivery teams — deployed functional code without fully grasping its internals. And it worked.

What AI Did Remarkably Well

The AI agents in this experiment performed at a level that genuinely impressed me. The code quality they generated was comparable to what I would expect from a competent front-end engineer. More notably, when I asked the agents to review their own outputs — essentially acting as a peer reviewer — they identified issues, proposed corrections, and iterated. The feedback loop was faster than any human sprint review cycle I have run.

For anyone who has spent years managing engineering teams, that observation carries significant weight. Code generation, syntax validation, self-review, and iteration — tasks that once required dedicated technical resources — are now accessible to a non-engineer willing to ask the right questions.

This experience highlighted several capabilities in the 2026 AI landscape:

• Code Quality: AI agents can produce front-end code at a level comparable to a skilled engineer, including peer review of their own outputs.
• Step-by-Step Guidance: AI walks non-technical users through complex setup processes, making previously gated tasks accessible.
• Rapid Iteration: Self-correction loops happen in seconds, compressing feedback cycles that traditionally took days.

The Lesson That Cut Deepest

The fact that I could deploy something I did not fully understand is not a testament to AI replacing expertise — it is a warning signal about the risks of AI-generated projects operating without proper oversight.

In a professional delivery context, "it works" is never the full acceptance criteria. We ask: Is it maintainable? Is it secure? Does it scale? Who owns it when it breaks at 2am? What is the rollback plan? These are not purely technical questions — they are governance, risk, and accountability questions. They are, in every meaningful sense, project management questions.

The more AI accelerates delivery, the more critical it becomes to have someone who can define what "done" actually looks like, ensure the right stakeholders have validated the output, and build a framework around AI-generated work that protects the organisation from the compounding risks of moving fast without sufficient understanding.

Where PM Skills Are More Essential Than Ever

My hands-on experience clarified something I have been sensing in conversations across the industry: AI-generated projects do not reduce the need for project management discipline — they amplify it.

Understanding the Risks of AI-Augmented Delivery

The experience of configuring a workflow I did not fully understand reveals a broader structural challenge in AI-assisted projects:

1. The Accountability Gap

AI generates outputs with confidence. It does not distinguish between a configuration that is functionally correct and one that is a security liability. Without a PM who understands the downstream implications of each decision, organisations can unknowingly accumulate technical and governance debt at machine speed.

2. The Prompt-to-Outcome Translation Problem

The quality of AI output is directly proportional to the clarity of the input. A vague business requirement fed into an AI agent will produce a precisely built solution to the wrong problem. This is not an AI failure — it is a requirements failure. And requirements have always been a PM's domain.

3. The Handover Challenge

When AI generates a system, who documents it? Who trains the team? Who owns it when the original builder moves on? These handover and knowledge-transfer challenges are intensified, not eliminated, when the code author is an AI agent rather than a human engineer.

Progress and Optimism: A PM's Perspective

Despite these challenges, the opportunity for project managers in this AI era is genuinely exciting. Teams that pair AI acceleration with strong PM discipline are achieving outcomes that were previously impossible within budget and time constraints.

Practical approaches for navigating AI-augmented project delivery include:

• Define Acceptance Criteria Before Prompting: Treat AI agent outputs the same way you would treat any vendor deliverable — with clear, pre-agreed acceptance criteria.
• Build a Review Layer: Even when AI peer-reviews its own code, schedule a human review milestone. AI is consistent; it is not infallible.
• Document AI-Generated Decisions: Maintain a decision log that records what the AI was asked, what it produced, and what human judgement was applied before sign-off.
• Engage Stakeholders Early: AI can prototype fast. Use that speed to generate conversation, not to skip the conversation entirely.

Reflections: The PM's Role in an AI World

This hands-on experiment with Activepieces and AI agents became an unexpected lesson in AI's current boundaries. While we celebrate the remarkable capabilities of generative AI — its ability to write, review, and iterate code at remarkable speed — we must also be clear-eyed about what it cannot do.

AI cannot own the outcome. It cannot navigate organisational politics. It cannot hold the vision when scope creeps at 11pm before a milestone. It cannot make the judgment call when two stakeholders have conflicting priorities. These capabilities are not incidental to project management — they are the heart of it.

My Challenge to Fellow PMs

Stop waiting for permission to engage with these tools. Run your own experiment. Break something, fix it, and pay attention to every moment where you instinctively reached for a risk log, a stakeholder conversation, or a governance framework — because those moments are where our value is crystallising in real time.

The profession is not at risk of disappearing. It is at risk of being left behind by PMs who refuse to understand the landscape they are being asked to navigate. The ones who lean in — who learn enough about AI to ask the hard questions, manage the risks, and own the outcomes — will be the most indispensable professionals in the room.

AI can generate the code. It cannot generate the judgment. That part is still very much ours.