Atlassian CEO breaks down the 3 types of SaaS in the AI era
1️⃣ Work-Tied Seats: High risk. If AI agents do the work, do you still need the seats?
2️⃣ System of Record: Safer. Pricing is not tied to output
3️⃣ The Middle Ground: Hybrid - balances creative work with tools
www.youtube.com/watch?v=0lzo...
"AI makes building cheap" is the most dangerous sentence in product.
Cheap to build ≠ worth building.
You can now create the wrong thing faster than ever.
Validation didn't get cheaper. It got more important.
#ProductManagement #AI
Decision log template that actually gets used:
1. What we decided
2. What alternatives did we considered
3. Why we chose this
Three lines. That's it.
6 months from now, nobody will remember the "why" unless you wrote it down.
#ProductManagement
Domino's didn't add GPS delivery because customers asked for it.
They noticed delivery notes kept saying "meet me at the beach" and "I'm in the park."
The best feature ideas hide in workarounds your users already invented.
#ProductManagement
Real question: Are you using AI to think better, or to avoid thinking?
If you can't write a clear problem statement without ChatGPT, that's not efficiency.
That's atrophy.
#ProductManagement #AI
The blank AI chat box is bad UX.
We spent decades perfecting form design, search patterns, and guided workflows.
Then AI shipped and said, "Just type anything" 🤣
No wonder people don't know what to do with it.
#AI #ProductManagement
Synthetic data is a PM superpower nobody talks about.
Use AI to generate test scenarios for edge cases, permission rules, and data APIs.
You don't need production data to validate your product logic.
#ProductManagement #AI
Pre-mortem > Post-mortem.
Before you build, ask your Eng and QA team: "If this feature fails, why did it fail?"
You'll catch integration gaps, edge cases, and wrong assumptions before they cost you a sprint.
#ProductManagement
70-75% AI-generated code is the new normal for MVPs.
The question isn't "is that too much?"
The question is, "Do you have someone who can tell good AI code from bad AI code?"
That's where the real risk lives.
#AI #ProductManagement
Stop planning your deliverables. Start planning your outcomes.
#ProductManagement
Built a personal PRM as an OpenClaw skill.
This one maps relationships as a graph, Obsidian-native, with weekly "who are you neglecting?" digests
Full write-up + open source skill.
Give it a try
gobiraj.substack.com...
#AI
In volatile markets, the winner is often the team that learns fastest, not the team that looks most polished.
#AI
Your strategy doc is more valuable for the thinking it forces than the document it produces.
#ProductManagement
I would say it is not binary but somewhere in the middle.AI won’t replace engineers, but it absolutely compresses execution cost. That changes team size, leverage, and expectations. There is a shift coming
Execution clarity is an underrated PM skill.
Especially now, when half your team is quietly anxious about AI replacing them.
You don't need a grand AI strategy.
You need to help each person on your team answer: "What does this mean for my work, this week?"
Before you build your roadmap, answer these first:
1️⃣ Who is your ICP?
2️⃣ What core problem are you solving?
3️⃣ What's your positioning?
Without these, your roadmap is just a wish list with dates.
#ProductManagement
The burning platform now isn't 'adopting AI.'
It's designing organizations that convert AI capability into reliable business outcomes.
That's the actual hard part nobody's solving.
#AI
AI Reality: Best users can also be the most expensive users.
#AI
You are right, not all use cases need an always-on agent. An agent on a recurring schedule should suffice
The hardest shift when building AI-native products isn't technical.
It's mental: going from SaaS-user mindset to infrastructure-operator mindset.
Flat-rate subscriptions weren't built for agents.
#AI
Old software economics: build it, ship it, collect recurring revenue.
New AI economics: constantly tuning prompts, forward-deployed engineering, monitoring outputs, and managing compute costs.
#AI
AI subscriptions were priced for humans. OpenClaw exposed the flaw.
PMs: your packaging should reflect real usage physics, not legacy assumptions.
#AI #ProductManagement
The real AI skill gap for PMs isn't prompt engineering.
It's knowing when AI output is "good enough" vs. confidently wrong.
AI doesn't know what it doesn't know. That's still a PM's job.
#ProductManagement #AI
The most useful thing you can add to an AI agent isn't better reasoning.
It's a cron job😀.
OpenClaw runs at 4am while you're asleep. You wake up, work is done.
Cron scheduling is the most underrated AI feature nobody's talking about.
#AI
The future isn't about making humans obsolete; it's about making them potent.
We're shifting from designing tools for humans to designing systems where humans & agents collaborate. The Oversight UI is the cockpit where that collaboration happens.
#ProductManagement #AI
6/6
4️⃣ Trust Builder: Trust requires transparency. The UI must provide an auditable "chain of reasoning" for every AI decision. Why was a $1,299 flight approved? The UI shows every policy check the agent made, turning a black box into a reliable system.
5/6
3️⃣ Escalation Point: The agent must know what it doesn't know. For an edge case like a blurry, hand-written receipt, the UI's job is to package all available data cleanly and escalate it to the right human expert for a manual decision.
4/6
2️⃣ Goal Refiner: The agent lacks context. If an expense is missing a project code, the UI shouldn't just error out. It should use context to suggest likely projects, turning a data-entry task into a one-click confirmation for the manager.
3/6
An Oversight UI is a collaborative cockpit for AI.
It has four key jobs.
Let's use an AI expense agent as an example.
1️⃣ Arbiter: When the AI faces a conflict (e.g., receipt vs. charge date), the UI presents evidence side-by-side for a human to make the final call.
2/6
Everyone's talking about AI agents replacing UI. They're missing the most important UX: The Oversight UI.
The popular narrative sees humans passively watching a dashboard. This is a myth. The real challenge isn't replacing the human; it's building their new cockpit.
1/6
