80% of the Workflow Was Automated With AI. Here’s What Happened Next.

Last updated: October 2025

Eight months ago, the goal became automating as much of the project-management workflow as possible using AI tools. Not as an experiment, but as a survival strategy after workload doubled following layoffs.

The automation worked. I got the evenings back. Then things got complicated.

What Got Automated

Meeting Notes and Action Items (Saved: 5 hours/week)

Before: Frantically typing during meetings, spending 20 minutes after each one cleaning up notes and sending action items.

After: Otter.ai records and transcribes every meeting. I paste the transcript into Claude with “Extract action items, decisions made, and open questions. Format as bullet points with owners and deadlines.” Takes 2 minutes per meeting instead of 20.

Quality check: The AI summaries are about 90% accurate. I scan them for errors before sending — usually fix 1-2 items per meeting. The 10% error rate sounds bad until you realize the manual notes had a similar error rate. I just didn’t notice because I never reviewed them.

Status Reports (Saved: 3 hours/week)

Before: Every Friday, I’d spend 2-3 hours compiling updates from Jira, Slack, and email into a status report for leadership.

After: the evaluation built a simple script that pulls data from Jira’s API (tickets completed, in progress, blocked) and feeds it to ChatGPT with a prompt template: “Write a project status update for leadership. Tone: professional but not corporate. Highlight risks. Be specific about blockers.”

The output needs light editing — maybe 15 minutes instead of 3 hours. The reports are actually better because the AI doesn’t forget to mention things I’d gloss over.

Email Drafting (Saved: 4 hours/week)

Before: Crafting careful emails to stakeholders, vendors, and team members. Agonizing over tone. Rewriting the same “just following up” message for the 400th time.

After: Claude drafts 80% of the emails. I describe the situation and the desired outcome, Claude writes the email, I review and send. For routine emails (meeting scheduling, status updates, follow-ups), I barely edit. For sensitive emails (escalations, bad news, negotiations), I rewrite heavily but use Claude’s draft as a starting point.

Jira Ticket Grooming (Saved: 2 hours/week)

Before: Writing detailed acceptance criteria, breaking epics into stories, estimating complexity.

After: I describe the feature in plain English, Claude generates acceptance criteria, suggests story breakdowns, and flags edge cases I might miss. The edge case detection alone is worth it — Claude consistently catches scenarios I’d overlook.

The Results: Month 1-3

The numbers were impressive:

• 14 hours/week saved on routine tasks
• Zero missed deadlines (previously missed 1-2 per quarter)
• Better report quality (leadership actually commented on this)
• Less overtime (from 50+ hours/week back to 40-42)

I felt like I’d discovered a cheat code. Same output, fraction of the time. the evaluation started taking on additional projects. Volunteered for cross-functional initiatives. My performance review was glowing.

The Complications: Month 4-6

The Visibility Problem

When you automate the visible parts of your job, people stop seeing you work. My status reports used to take all Friday afternoon — everyone knew I was working on them. Now they appeared in 20 minutes. My meeting summaries used to come out hours later, clearly hand-crafted. Now they arrived within minutes, suspiciously polished.

Nobody said anything directly. But analysis revealed a shift. In a planning meeting, the director asked, “Do you have capacity for more projects?” Not because I was doing less — I was doing more. But the effort was invisible.

The Skill Atrophy Question

After 4 months of AI-drafted emails, analysis revealed something uncomfortable: my first drafts without AI were worse. Not dramatically — but the muscle memory of crafting a persuasive email from scratch had weakened. I was out of practice.

This scared me enough to start writing one email per day completely manually. Like going to the gym for a skill The evaluation used to have naturally.

The Trust Tax

My team figured out I was using AI. Not because the output was bad — because it was too consistent. Humans have off days. AI doesn’t. When every meeting summary is perfectly structured and every status report hits the same quality bar, people notice.

Some team members were fine with it. Others felt weird that their words were being processed by AI. I had to have conversations about data handling and privacy that I hadn’t anticipated.

The Adjustment: Month 6-8

I landed on a sustainable approach:

1. Automate the invisible stuff. Data compilation, first drafts, template-based work. Nobody cares how the sausage is made.
2. Keep the human touch on visible stuff. I now manually write the opening paragraph of status reports and add personal observations. The AI does the data section. Best of both worlds.
3. Be transparent. I told the team and the manager that I use AI tools. Framed it as “I use AI to handle administrative work so I can focus on strategy and people.” Nobody objected.
4. Maintain the skills. One fully manual email per day. One meeting where I take notes by hand. Enough to keep the muscles working.

What I Learned

AI doesn’t replace your job. It replaces the parts of your job that aren’t really your job. Meeting notes aren’t project management. Status reports aren’t leadership. Email isn’t strategy. Automating the administrative layer revealed that the actual value I provide — decision-making, relationship management, risk assessment, team motivation — was always there. It was just buried under busywork.

The 80% automation number is misleading. I automated 80% of the tasks by volume, but those tasks represented maybe 40% of the actual value. The remaining 20% of tasks — the ones AI can’t do — are the ones that matter most.

Transparency beats secrecy. Hiding AI use creates anxiety (yours and others’). Being open about it positions you as innovative rather than lazy.

The goal isn’t to work less. It’s to work on the right things. I didn’t use the saved 14 hours to leave early (mostly). The evaluation used them to do the strategic work I never had time for — process improvements, team development, stakeholder relationships. The stuff that gets you promoted.

Current AI Stack for Work

$45/month to save 14 hours/week. At any professional salary, that’s the best ROI you’ll find.

For specific tool recommendations by role, check our guides for project managers, sales teams, and HR professionals.