DEV Community: Marcus

I Asked AI to Audit My Last 3 Failed Projects — Here's What It Found

Marcus — Mon, 27 Apr 2026 17:11:55 +0000

Two years ago, I was managing three concurrent projects. All three ran over budget. Two missed their deadlines. One got cancelled.

I blamed external factors at the time: changing requirements, staffing issues, unclear leadership direction.

Last month, I fed all three post-mortems into an AI and asked it to find the patterns. What it found was uncomfortable.

The Experiment

I had post-mortems from three projects that went sideways. Each was 2-4 pages of honest documentation.

I gave them to an AI with this prompt:

Here are three project post-mortems from the same project manager. 
Read them and identify:
1. Patterns that appear across all three failures
2. Root causes vs. symptoms
3. Early warning signals that appeared in all three
4. Be honest. I want genuine insight, not validation.

What It Found

Pattern 1: I was confusing status with progress.

All three post-mortems mentioned 'good communication' and 'regular check-ins.' The AI noted that while I had excellent visibility into WHERE the projects were, I consistently lacked early clarity on whether the PACE was sustainable. Lots of green status updates that turned red in weeks 8-9.

I thought I was managing. I was reporting.

Pattern 2: Scope was agreed but never defined.

In all three cases, stakeholders had signed off on scope. But the documentation left room for interpretation — phrases like 'a basic version of X' or 'initial integration with Y.' When reality hit, everyone had a different mental model of what 'basic' meant.

I knew scope creep was a risk. I didn't realize I had built ambiguity into the scope itself.

Pattern 3: The first 4 weeks set the ceiling.

The AI pointed out that decisions made in weeks 1-4 constrained what was possible for the rest of the timeline. Architecture choices, stakeholder alignment decisions, and resource allocations made early became increasingly hard to reverse.

'Your post-mortems consistently describe problems in weeks 8-12 that were seeded in weeks 1-4.'

What I Changed

After this analysis, I made three concrete changes:

1. Replace status reports with pace reports. Instead of 'here's where we are,' I now include 'here's whether we're on trajectory to hit the deadline at current pace.'

2. Define scope with examples of what's NOT included. For every major scope item, I document one or two things that might seem included but aren't. Explicit exclusions prevent the most common scope debates.

3. The Week 4 review. I now run a mandatory retrospective at week 4, specifically asking: what decisions made so far will be hardest to undo? Are we comfortable with those?

The Uncomfortable Part

What made this exercise valuable was that AI doesn't have a stake in protecting your ego. When you debrief with colleagues, there's social friction around saying 'you consistently made the same mistake.' With AI, there's no friction. It just says the thing.

If you're interested in using AI as a genuine thought partner in project management — not just a writing assistant — I write a free weekly newsletter about exactly this:

Ahead of Schedule: practical AI habits for project managers. One thing you can use this week, every week.

Subscribe free: https://buttondown.com/marcustillerman

Why Your Projects Keep Getting Derailed (And How AI Can Fix It)

Marcus — Mon, 27 Apr 2026 17:07:53 +0000

I've been managing software projects for over a decade, and I've noticed a pattern: most project failures aren't caused by the problems you anticipated. They're caused by the ones you didn't.

The risks you named in your risk register rarely derail you. The blindspots do.

Here's how I use AI to surface blindspots before they become disasters.

Why Traditional Risk Management Fails

Most project risk registers look like this:

Risk	Likelihood	Impact	Mitigation
Key person leaves	Low	High	Document everything
Budget overrun	Medium	High	Track weekly

This table exists in a spreadsheet nobody reads until something goes wrong. And the risks on it are the obvious ones — the risks that anyone who's been burned before would put down.

The dangerous risks are the ones you didn't think to add.

The AI Blindspot Finder

Here's the prompt I run at the start of every new project:

I'm managing a [describe project: industry, team size, timeline, technology, 
key stakeholders, what success looks like]. 

Give me:
1. The 5 most common failure modes for projects like this
2. The 5 LESS OBVIOUS risks that people in this situation typically miss
3. For each risk: what does the early warning signal look like?
4. Which 2-3 risks, if they hit simultaneously, would be catastrophic?

Be specific to the context I've described, not generic.

The key is that last instruction: be specific, not generic.

A generic prompt returns generic risks. A prompt with real context returns risks that are actually relevant to your project.

What Good Output Looks Like

For a recent project (new B2B SaaS feature, 6-person team, 3-month timeline, integrating with a legacy API), the AI surfaced:

Common failure modes (expected):

Scope creep from unclear requirements
API integration taking 2x longer than estimated

Less obvious risks (valuable):

"Integration-first" scope trap: Teams sometimes optimize too much around the limitations of the legacy API instead of building the right product, then retrofitting
Stakeholder misalignment on definition of "done" — B2B features often have different success criteria for different buyers (end user vs. procurement vs. IT)
Estimated 3 months is aggressive for first integration with this type of API. Teams often underestimate the documentation gap (docs say X, actual behavior is Y)

That third point led to a direct conversation with the legacy API vendor in week 1, which saved us from a nasty surprise in week 8.

The Weekly Risk Pulse

Running a risk check at kickoff is table stakes. The real value is making it a weekly habit.

I keep a running project context doc (3-4 sentences on current status, recent decisions, emerging issues). Every Monday:

Project context: [paste current status summary]
New developments this week: [list any decisions, changes, blockers]

Given this update, what risks have increased in probability? 
Are there any new risks that weren't relevant before but are now?
What should I be watching for this week?

This takes 5 minutes and has caught at least 3 significant problems before they became crises.

The Compound Risk Analysis

The most underused type of risk analysis is asking: what if multiple things go wrong at once?

Here are the top 5 risks on my project: [list]. 
Which combinations of these would be most devastating if they occurred 
simultaneously? For each combination, what's the trigger point where I 
should escalate, and what would the recovery plan look like?

The value isn't just the answer — it's that thinking through compound failures forces you to separate recoverable situations from catastrophic ones, which changes how you prioritize mitigation.

Bottom Line

The best thing AI does for risk management isn't building the risk register faster. It's helping you see the risks you would have missed entirely.

The PM's job becomes: give AI good context, evaluate the output with your judgment, and act on the things you couldn't see before.

If you found this useful, I write a free weekly newsletter called Ahead of Schedule about practical AI habits for project managers.

One thing you can use this week, every week: https://buttondown.com/marcustillerman

How I Reduced Meeting Time by 30% Using This One AI Habit

Marcus — Mon, 27 Apr 2026 17:00:26 +0000

Every week I was spending 8-10 hours in meetings. Most of them were fine. But a significant portion were meetings that either didn't need to happen at all, or ran 30 minutes longer than they should have.

After some digging, I realized the root cause wasn't discipline or culture — it was a preparation problem. Most meetings failed before they started, because the agenda was unclear.

Here's the AI habit that changed that.

The Problem with Most Meeting Agendas

A typical agenda looks like this:

Project status update (30 min)
Q3 planning discussion (20 min)
Misc / AOB (10 min)

This agenda is useless. It doesn't tell anyone what decisions need to be made, who is responsible for each item, or what success looks like. It's a list of topics masquerading as a plan.

The result: meetings where people talk until time runs out, then schedule a follow-up.

The Habit: AI Agenda Review, Every Time

Before every meeting I organize or attend, I run the agenda through this prompt:

Here's my meeting agenda: [paste agenda].

Review it and:
1. Flag any item that doesn't have a clear decision or outcome stated
2. Suggest which items could be handled async instead
3. Add a time box to each remaining item
4. Identify who should own driving each agenda item
5. Suggest one clarifying question I should ask before the meeting starts

Meeting details: [duration, attendees/roles, context]

What It Catches

Here's what this catches that humans routinely miss:

"Status update" items — 80% of the time, these can be a Slack message or email. AI will flag these consistently.

Decision items without a decision-maker — If there's no clear accountable person, you're setting up for a circular discussion.

Bloated agendas — A 60-minute meeting with 8 agenda items is a fantasy. AI will do the time math and tell you which items to cut.

Missing pre-work — Sometimes the right outcome is "get data from X before we can decide." AI catches when you're trying to make decisions without the necessary information.

Real Example

I recently had a sprint review agenda:

Demo new features (15 min)
Discuss sprint velocity (10 min)
Backlog grooming (20 min)
Team feedback (10 min)
Planning for next sprint (15 min)

Running it through the prompt revealed:

"Discuss sprint velocity" could be an async Slack post + reactions
"Team feedback" had no owner or specific questions
I was trying to do backlog grooming AND sprint planning in 35 minutes — impossible

The revised meeting was 45 minutes instead of 70, and we actually finished it.

The Bigger Picture

The goal isn't just shorter meetings — it's meetings where everyone knows why they're there and what they need to contribute.

When you arrive at a meeting and the agenda is crystal clear, you can prepare. When you prepare, you contribute better. When everyone contributes better, decisions happen faster.

That's how you get 30% of your meeting time back.

I write a weekly newsletter called Ahead of Schedule about practical AI habits for project managers and team leads. Every issue is one thing you can use this week.

Subscribe free: https://buttondown.com/marcustillerman

What's your biggest meeting pain point? Drop it in the comments — I'll cover it in a future issue.

5 AI Prompts That Saved My PM Team 40+ Hours Per Month

Marcus — Mon, 27 Apr 2026 16:54:05 +0000

I'm a Senior PM at a software company, and I was drowning in status reports, kickoff docs, and uncomfortable conversations.

Then I started treating AI as a thought partner rather than a writing tool — and everything changed.

Here are the 5 prompts that genuinely transformed how I work. I'm sharing these publicly because I think most PMs are massively underusing AI.

Prompt 1: The Stakeholder Update Generator

The problem: 45 minutes every Friday writing a status update nobody fully reads.

I need to write a weekly stakeholder update for [project name]. 
Here is the raw data: [paste your notes/Jira summary/Slack messages]. 
Write a 3-paragraph update that starts with the most important news, 
explains any risks or blockers clearly, and ends with what is needed 
from stakeholders. Keep it under 200 words and avoid technical jargon.

Why it works: AI synthesizes messy input into clean narrative. Paste in notes, get a polished draft in 30 seconds, edit for 5 minutes.

Time saved: ~40 min/week

Prompt 2: The Meeting Agenda Optimizer

The problem: Meetings run long because the agenda is vague.

Here's my meeting agenda for a 60-minute [type] meeting: [paste agenda]. 
Rewrite this so every item has: a clear decision or outcome needed, 
a time box, and the one person responsible. Flag any items that could 
be an async update instead.

Why it works: Often, agenda items are FYI updates that did not need a meeting at all.

Time saved: 20-30 min of wasted meeting time per session

Prompt 3: The Risk Surface Finder

The problem: The risks you do not see are the ones that derail projects.

I'm managing a [describe project: type, team size, timeline, tech stack]. 
What are the top 10 risks I should be tracking? For each: plain-English 
description, likelihood (H/M/L), impact, and one early warning sign.

Why it works: AI has absorbed patterns from thousands of similar projects. It surfaces risks you would not think of until too late.

Time saved: 2-3 hours at project kickoff

Prompt 4: The Difficult Conversation Prep

The problem: You need to tell a stakeholder their feature is getting cut.

I need to have a difficult conversation with [their role] about [the issue]. 
Key facts: [facts]. I want to come across as direct but respectful, focused 
on solutions not blame. Write me an opening 2-3 sentences and suggest 3 ways 
they might respond and how I should handle each.

Why it works: Roleplay prep with AI is underrated. Seeing likely responses makes you far less anxious going in.

Prompt 5: The Scope Creep Detector

The problem: Scope creep sneaks in through small asks that individually seem fine.

Here is the original scope: [paste scope/SOW]. Here are the requests that 
have come in since kickoff: [list]. For each: (1) does it fall within 
original scope? (2) complexity estimate? (3) what should I say to the 
stakeholder who asked?

Why it works: Turns a judgment call into explicit analysis. The scope conversation is easier when you have thought it through first.

The Bigger Picture

These prompts save me 3-4 hours per week. More importantly, they let me focus on the actual work of project management: alignment, decision-making, and unblocking my team.

If you are a PM wanting to go deeper on this, I write a free weekly newsletter called Ahead of Schedule — practical AI tools and strategies specifically for project managers. One thing you can use this week, every week.

Subscribe at: https://buttondown.com/marcustillerman

What prompts are you using? Drop them in the comments.