AI engineering leader says dedicated project management roles will disappear
See how an AI engineering leader uses agentic workflows to speed up delivery, reduce meetings, and shift teams toward judgment-focused collaboration.
Sairam Sundaresan
AI Engineering Leader and Educator
AI Impact: AI significantly alters the focus of AI/ML engineering roles, shifting emphasis to design and problem framing.
Task Automation: AI excels in automating repetitive tasks but struggles with tasks requiring judgment and context understanding.
Delivery Rituals: AI facilitates faster project scoping and enhances asynchronous alignment, reducing the need for status update meetings.
Agentic Workflows: Implementing agentic workflows streamlines non-coding tasks, allowing for better focus on critical decision-making.
Role Evolution: Dedicated project management roles may diminish as AI takes over tracking, leaving judgment-based tasks to teams.
Sairam Sundaresan is an AI engineering leader and educator. He's ranked as the #1 AI Creator on LinkedIn India by Favikon. And he's the author of AI for the Rest of Us and a weekly newsletter called Gradient Ascent.
We spoke to Sundaresan about how AI is changing project delivery. Here is what he had to say.
How AI influences AI/ML engineering roles
I lead AI/ML engineering. In project delivery, I'm responsible for architecture decisions, making sure the technical approach maps to the business problem, and catching risks early enough to address them.
Unlock for Free
Create a free account to finish this piece and join a community of forward-thinking leaders unlocking tools, playbooks, and insights for thriving in the age of AI.
Have an account? Log In
I'm also the author of AI for the Rest of Us.
How AI shifts project delivery tasks and focus
AI does a good first pass on implementation, tests, and boilerplate. I spend less time writing code from scratch, but I still review everything, the same as I would with any engineer on my team. The output is produced faster, but it's unsupervised.
I now spend more time on problem framing, system design, and figuring out where things will break. Those decisions were always the high-leverage part of the job. They take up a larger share of my day now that routine work moves faster.
What surprised me most about implementing AI was Moravec's paradox playing out in real time. The tasks I expected to be hard to automate were easy. Code generation, test writing, and summarization. AI struggles with tasks I expected to be trivial: knowing when a requirement is ambiguous, recognizing that a technically correct solution is the wrong one for this context, and understanding why a stakeholder said one thing but meant another.
How AI optimizes repetitive tasks in project delivery
The most automatable parts of my delivery work right now are repetitive tasks with clear acceptance criteria. Test generation, code scaffolding, writing first drafts of documentation, status reporting, and dependency checks — these are well-defined, low-ambiguity tasks where AI does a solid job with minimal oversight. Implementation is straightforward: you plug agents into your CI/CD pipeline, give them a clear scope, and build verification around the output.
Humans are still needed for anything involving judgment under ambiguity. That includes:
- Prioritization when everything is urgent
- Knowing which technical shortcut is acceptable and which one will cost you six months later
- Reading the room in a cross-functional meeting and realizing the real blocker is political.
How AI reshapes core delivery rituals and alignment
As far as rituals, AI does a lot of the heavy lifting.
- Scoping is faster because AI generates a first draft from a conversation or requirements, and then I edit. I'm not staring at a blank page anymore. But I still make the judgment call on what's in scope and what's not.
- Alignment is more async. Instead of a meeting to align everyone, I write a short doc, AI helps me sharpen it, and people comment. We meet only when we need to resolve a genuine disagreement.
- Validation changed the most. AI generates code, so now I need tests and guardrails that catch agents' specific failure modes.
- Execution management is lighter. Tools provide status directly. I spend less time asking for status updates and more time deciding what to do next.
Alignment is more async. Instead of a meeting to align everyone, I write a short doc, AI helps me sharpen it, and people comment. We meet only when we need to resolve a genuine disagreement.
Why agentic workflows enhance delivery efficiency
Personally, I use agentic workflows daily. My setup uses Claude Code to orchestrate most of my non-coding work: research, document synthesis, planning, and review. For coding, Codex handles the implementation, and I review.
I first focused on automating aspects of delivery that consumed time without requiring judgment. Drafting docs, summarizing context, generating tests, and pulling together status from multiple sources. These are high-frequency, low-ambiguity tasks. Easy wins.
The quality is good enough that I only edit the work, not redo it. I use the time I get back for decisions that actually matter. The main lesson: be specific about scope. Agents do well with clear boundaries. Give them ambiguous intent, and you're back to babysitting.
Why Codex and Claude Code are must-have dev tools
Codex is my primary coding tool right now. When I use Claude for coding, Codex acts as my code-review partner. However, this role changes as the space evolves.
Claude Code handles everything related to coding: architecture thinking, documentation, problem framing, research, and review. It also performs some direct coding. I use Linear for project tracking. It is clean, fast, and stays out of the way.
Why AI may eliminate dedicated project management roles
This may be a bit controversial, but I predict dedicated project management roles will mostly go away because everyone will start managing projects. AI handles tracking, status rollups, and scheduling. What's left is judgment, and the team shares that.
All roles are blending right now. You can't be just an engineer who codes, or just a designer who designs. Smaller teams, more AI leverage, and everyone will cover more surface area.
The people who do well are the ones who were never just one thing.
Why delivery leaders must embrace AI tools and skills
Get your hands dirty with the tools. Don't delegate your understanding of AI to someone else on your team. You need to feel where it's good and where it breaks, firsthand.
After that, ruthlessly cut processes that exist only because information was expensive to move around. AI gives you that for free now.
Follow Along
You can follow Sundaresan's work on LinkedIn, X, Instagram, and his newsletter. You can also check out his book, AI for the Rest of Us.
More expert interviews to come on The Digital Project Manager!
