Most project managers I talk to use AI the same way they use Google: they ask a question, get an answer, and move on.
\nThat’s not AI-augmented project management. That’s a slightly faster search engine.
\nThe real unlock \u2014 the one that changed how I run projects at Capgemini and how I built Bartronics from zero to 120 live bots \u2014 is treating AI as a team member, not a tool. Specifically, as an agent: something that takes a task, runs with it, and brings back a deliverable you can actually use.
\nHere’s exactly how I do it.<\/p>\n
An AI agent, in practical terms, is a prompt-engineered workflow where you give the AI a role, context, constraints, and a deliverable \u2014 and it produces something ready to use with minimal back-and-forth.
\nNot: “What are the risks of a software integration project?”
\nBut: “You are a senior risk manager. I am the PM for a B2B API integration between a ticketing system and 3 enterprise clients across Colombia, Mexico, and Spain. Timeline is 12 weeks. Identify the top 10 risks, score each by probability (1-5) and impact (1-5), and produce a risk register table with mitigation strategies.”
\nThe second prompt gets you a document. The first gets you a list.<\/p>\n
This one saves me 20-30 minutes after every stakeholder meeting. I paste the raw notes \u2014 even messy, unstructured ones \u2014 and use this prompt:
\n“You are a project manager assistant. Below are raw notes from a stakeholder meeting. Extract all action items, assign an owner based on context, set a reasonable due date assuming today is [date], and format as a table with columns: Action Item | Owner | Due Date | Priority (High\/Medium\/Low).”
\nThe output goes directly into Monday.com or Jira. No reformatting needed.<\/p>\n
Before every project kickoff I generate a first-draft risk register in under 5 minutes. The prompt includes the project type, industry, team size, and timeline. Claude or ChatGPT returns a structured table with 10-15 risks, probability\/impact scores, and mitigation strategies.
\nI review it, add domain-specific risks I know from experience, and present it to the client. What used to take half a day now takes 20 minutes.<\/p>\n
Escalation emails, status updates, difficult conversations \u2014 I draft them with AI first. The key is giving it the full context: who the stakeholder is, what the issue is, what outcome I want, and what tone is appropriate.
\nFor a delayed delivery notification to a C-level at Cisco, for example, I’d write: “Draft an email to a VP of IT at Cisco informing them that the API integration milestone is delayed by 2 weeks due to incomplete documentation from a third-party vendor. Tone: professional, direct, solution-focused. Include the revised timeline and what we are doing to recover.”
\nThe draft is 80% ready. I personalize the last 20%.<\/p>\n
I use AI to stress-test roadmaps before presenting them. I paste the proposed timeline and ask: “Review this project roadmap and identify unrealistic assumptions, missing dependencies, and risks to the timeline. Suggest adjustments.”
\nIt catches things I miss. Not because AI knows more than me \u2014 but because it reads the document without the cognitive bias of the person who built it.<\/p>\n
At Bartronics, where I lead a team of 4, I use AI to structure quarterly skill gap analysis. I describe the team’s current skills and the upcoming project requirements, and ask for a gap analysis with learning recommendations per person.
\nThis scales \u2014 the same workflow works for a team of 4 or 40.<\/p>\n
When we scaled from the initial automation bots to building our own VPS infrastructure and then the WhatsApp Business API service, I was managing 3-4 concurrent projects with a lean team.
\nAI agents became the force multiplier. Every project had:<\/p>\n
An AI-generated risk register updated weekly
\nAutomated meeting summaries converted to action items
\nRoadmap drafts stress-tested before stakeholder presentations
\nStatus report templates filled with project data in minutes<\/p>\n
The result: I managed a program that reached ~$6K USD MRR in 9 months without burning out or missing deliveries. Not because I’m superhuman \u2014 because I offloaded the cognitive overhead of documentation to AI and kept my focus on decisions, relationships, and strategy.<\/p>\n
I use three AI tools in rotation, and each has a different strength:
\nChatGPT (GPT-4o) \u2014 best for structured document generation. Risk registers, project charters, meeting agendas. Follows templates precisely.
\nClaude (Anthropic) \u2014 best for nuanced writing and analysis. Stakeholder communications, complex email drafts, roadmap reviews. Better at tone.
\nGemini \u2014 best when I need web context integrated. Researching industry benchmarks, competitor analysis for a client presentation, or pulling recent data into a report.
\nI don’t pick one and commit. I use whichever is best for the specific deliverable.<\/p>\n
AI doesn’t replace the judgment call. It doesn’t tell you whether to escalate a conflict with a stakeholder or manage it quietly. It doesn’t know that the client’s VP hates slide decks and prefers a 5-minute call. It doesn’t have the relationship context that makes or breaks a project.
\nThose are the things a senior PM does. AI handles the paperwork so you can focus on them.
\nThe PMs who will be replaced by AI are not the ones who use it \u2014 they’re the ones who spend 80% of their time on documentation and administrative tasks that AI can now do in minutes.<\/p>\n
If you’re new to this, start with one workflow: action items from meeting notes. It’s the highest-frequency, lowest-risk place to integrate AI. After 2 weeks it becomes automatic, and you’ll start seeing where else it fits.
\nThe prompt template is above. Paste your next meeting notes and try it.<\/p>\n
Alejandro Barahona is a PMP\u00ae-certified Senior Project Manager specializing in software integrations and automation. He currently leads B2B API integration projects at Capgemini Engineering and is co-founder of Bartronics. Available for remote roles \u2014 connect on WhatsApp or view portfolio.<\/p>\n","protected":false},"excerpt":{"rendered":"
Most project managers I talk to use AI the same way they use Google: they ask a question, get an answer, and move on. That’s not AI-augmented project\u2026<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[8,14,11,13,10,12,9],"class_list":["post-9","post","type-post","status-publish","format-standard","hentry","category-ai-project-management","tag-ai","tag-artificial-intelligence","tag-automation","tag-chatgpt","tag-claude","tag-pm-productivity","tag-project-management"],"yoast_head":"\r\n