Product Management in the AI World Is Like Playing StarCraft

Or any other Real-Time Strategy game (AoE2, C&C, anyone?)

Growing up, I spent an embarrassing number of middle and high school afternoons watching StarCraft replays. Not playing, but watching Artosis casting a Flash vs. Jaedong series to try to understand what I was seeing. It took a while before I could even follow it. The actions were too fast, the decisions too compressed.

But eventually something clicked. You stop watching the battle and start watching the player. The camera flicks across the map (base, expansion, army, back to base) in a rhythm so precise it almost looks choreographed. No wasted motion. No dead time. Everything happening at once, everything connected, and soon, you understood things like “build orders”, strategies to use against Terran vs Zerg vs Protoss (I was somewhat of a Toss player, myself).

You start realizing that the players aren’t reacting to things as they experienced them; they were just executing a system with high intensity.

I’ve been thinking about those replays a lot lately in the context of product management because the best PMs I’ve worked with operate the same way. And with AI tooling becoming genuinely good, I think we’re on the verge of a step-change in what “great PM execution” even looks like.

01 · The Game Within the Game

If you haven’t played StarCraft: Brood War, or watched it at a high level, here’s the core tension: the game demands that you operate across two completely different scales simultaneously.

Macro is your long game. Economy. Build order. When to expand, when to tech up, when to time an attack. Macro is about resource allocation, strategic sequencing, and reading where the game is going before it gets there. A player with great macro is never supply-blocked, never floating minerals, never caught without units when it matters.

Micro is the moment-to-moment. How you micro your SCVs against a Zergling rush. How Jaedong moves his Mutalisk flock to avoid Psy Storms while simultaneously harassing three bases. How Bisu’s Protoss Dragoons dance around a choke point without losing a single unit. Micro is execution under pressure, the fine motor skill of turning good pieces into great outcomes.

The brutal truth of StarCraft is that you need both. Brilliant micro with no macro means you’re fighting beautiful losing battles. Pristine macro with no micro means you build a perfect army and then watch it walk into a bunker line.

Great product managers are also playing two games at once, and most frameworks only ever teach you one of them.

02 · Your Build Order Is Your Roadmap

PM macro is everything that shapes the long arc of the product. Roadmap prioritization. Bet sizing. Sequencing features so they compound rather than conflict. Understanding your resource constraints (engineering capacity, design bandwidth, your own time) well enough to never be “supply blocked” going into a planning cycle.

It’s also market reading. In StarCraft, a player who scouts early and adapts their build order has a decisive information advantage. In product, that’s the PM who’s talking to customers every week, watching usage data like a hawk, and adjusting the roadmap before the quarterly review forces it. At Amplitude, we live and die by this. When your product is a product analytics platform, you have no excuse for not knowing exactly where users are dropping off, which features are sticky, and what the data says before any planning conversation starts.

The translation maps cleanly:

Build order is your roadmap sequencing and dependency management.
Economy and worker production is headcount planning and sprint capacity.
Scouting and map awareness is win/loss analysis, customer discovery, and competitive intel.
Expansion timing is knowing when to enter a new segment or bet on a new surface area.
Tech tree decisions are your platform vs. feature tradeoffs and build vs. buy calls.

A PM with strong macro ships predictably. They almost never surprise their stakeholders, not because they lack ambition, but because they’ve already socialized the hard calls and sequenced the work so it lands cleanly. Their planning cycles feel calm. Their teams trust the north star, and in turn, their teams trust them.

03 · Micro Is Where the Game Is Actually Won

This is the part most PM content skips. Macro is the sexy framework. Micro is the unglamorous grind that separates okay PMs from great ones.

PM micro is everything at the execution layer. How you write a spec: is it clear enough that engineering doesn’t need a follow-up meeting? How you run a discovery call: are you actually uncovering the problem beneath the problem, or just validating your assumptions? How you prep for a stakeholder review: do you anticipate the objection before the slide lands?

The same mapping holds at the micro level:

Unit control under fire is running a crisp incident postmortem while simultaneously shipping the fix.
Mutalisk harassment is unblocking an eng dependency before it ever shows up on the board.
Stutter-stepping is moving a stakeholder from skeptic to advocate through communication and storytelling.
Skill timing (storm, EMP, Lockdown) is knowing exactly when to escalate versus absorb pressure yourself.
Splitting against splash damage is breaking up a PRD so teams can work in parallel without stepping on each other.

Micro is also the interpersonal game. The best PMs are doing constant unit control on their relationships, knowing when to push, when to give ground, when to pull someone in versus route around them. That’s not manipulation. That’s craft.

04 · What APM Actually Measures

Actions Per Minute in StarCraft isn’t just speed. It’s a proxy for how much of the game a player can hold in their head and act on simultaneously. A player at 80 APM is probably reacting. A player at 200+ APM is running routines: automatic worker production, supply depot placement, camera hotkeys firing in a rhythm while they’re thinking three steps ahead.

The legendary players (Jaedong, Bisu, Flash, and my personal favorite: Reach) weren’t fast because they were frantic. They were fast because they had automated so much of the cognitive overhead that their full attention could live at the strategic layer.

This is the real lesson for PMs. Our version of APM isn’t how many Slack messages we send. It’s how many meaningful decisions and outputs we can move through in a day (specs written, discovery calls run, stakeholder alignments landed, data pulls synthesized) while our strategic brain stays free to think about the big picture.

Most PMs are playing at 80 APM. Not because they’re slow, but because they haven’t automated the routine yet.

05 · AI Is the Hotkey Revolution

Here’s the thing about the evolution of StarCraft at a high level: APM didn’t go up because players got biologically faster. It went up because players developed systems. Hotkey groups. Camera anchors. Muscle-memory build orders. The cognitive overhead of the routine got compressed until it was nearly zero, freeing up bandwidth for everything else.

That’s exactly what’s happening with AI tooling right now for PMs. And I think most people are underestimating the magnitude of the shift.

Think about where your PM cognitive load actually goes on a given week: roughly 60% of PM time goes to synthesis, writing, and communication. Not the actual thinking. Research synthesis. Spec writing. Slide decks. Meeting prep. Status updates. Jira tickets. These aren’t the hard part of the job, but they consume the most time. And more insidiously, they fragment the deep thinking that macro strategy requires.

AI revolutionizes this.

When I can turn a rough brain dump into a clean PRD in 20 minutes instead of 3 hours, I have 2.5 hours back. When I can synthesize 40 customer interviews into a structured insight map in an afternoon instead of a week, my discovery velocity doubles. When I can draft the stakeholder deck while I’m already in the next problem, I’m not context-switching. I’m parallelizing. Running monetization strategy at Amplitude, that means I can be in the data modeling for a new entitlement tier while an agent is already summarizing the latest win/loss calls from the self-serve segment. Two strategic threads, one working session. And if you’re a technical PM working day-in and day-out with your eng team, you can even ship full-stack features to accelerate and unblock the highest priority bets.

That’s APM going from 80 to 300.

05b · The Build: Agents, Skills, and MCP

The real unlock isn’t just using AI as a smarter search box. It’s wiring it into your actual workflow so the routine runs itself. Here’s what that looks like in practice with modern tooling.

Claude Agents are the equivalent of having a player run automated build routines in the background while you focus on the army. Instead of switching context to check on a task, an agent handles it and reports back. A PM might spin up an agent to monitor a competitor’s changelog weekly, pull the diffs, and surface anything that affects your roadmap. It runs while you’re in your sprint review. You get the brief after.

Skills are reusable instruction sets you build once and deploy every time a recurring workflow comes up. Think of them as your saved hotkey configurations. You write a PRD skill that knows your team’s spec template, your company’s product principles, and the level of detail your engineering team needs. From that point on, every PRD you generate starts from that foundation, not a blank page. You can build skills for customer interview synthesis, competitive teardowns, OKR drafts, release notes. Each one compounds. The more you build, the faster the routine gets.

MCP (Model Context Protocol) is the integration layer that connects your agent to the tools where work actually lives. Your agent can read your Jira board, pull usage data from your analytics platform, fetch the latest customer feedback from your CRM, and cross-reference it all in a single workflow. No manual context switching. No copy-pasting between tabs. One prompt, one output, sourced from every relevant system. Specific examples that are already available: GitHub MCP for surfacing engineering commit history when writing release notes, Notion MCP for meeting prep that auto-pulls the right internal docs, Amplitude MCP for answering product analytics questions directly in your workflow without waiting on a data pull. If you’re working on self-serve growth or monetization, being able to ask “what’s the conversion rate from free to paid for users who hit the collaboration limit” and get an answer in seconds rather than filing a data request changes the entire pace of decision-making.

Put them together and you get something genuinely different. An agent running a competitive analysis skill, pulling live data from Google Drive, your internal docs, and the web via MCP, formatted to your team’s standards, ready before your Monday planning call. That’s not an AI assistant. That’s a system.

In Brood War terms: you’re not pressing keys faster. You’re building the macro engine so the economy runs on autopilot while your attention stays on the strategic engagements that actually win the game.

06 · But the Fundamentals Don’t Compress

Here’s the important caveat, and the reason this analogy holds all the way down.

AI tooling accelerates your execution. It does not improve your judgment. A PM who doesn’t understand why a feature matters to customers will generate a beautiful, coherent, wrong PRD faster than ever. A PM who can’t read their stakeholder landscape will prep the perfect slide deck for a meeting they were always going to lose.

In StarCraft, a player with 300 APM and no strategic sense will just lose faster. Execution speed amplifies your fundamentals. It doesn’t replace them. You don’t go all-in Mutas into a Toss player with Corsairs on the field (unless you’re hella badass).

Jaedong’s Zerg control was insane, but it was extraordinary in service of a strategy. The micro existed to execute the macro.

The PMs who will get the most from AI are the ones who have already done the work to develop strong judgment: they know what problems are worth solving, they can read a room, they understand the engineering tradeoffs, they’ve built the institutional trust to move fast. AI gives those PMs leverage.

07 · What Playing at 300 APM Looks Like

I want to be concrete about this, because “AI makes you faster” is not enough.

Discovery at speed. Run 3x the customer interviews because you’re prepping faster, synthesizing faster, and generating follow-up hypotheses between calls instead of two days later. Your discovery cycles compress from months to weeks.

Always-current specs. Your PRDs don’t go stale because the cost of updating them dropped to near zero. Engineering never works from an outdated document. Alignment is continuous, not point-in-time.

Proactive stakeholder coverage. You’re generating the status update, the risk flag, the pre-read, the post-meeting summary. Not because you have more time, but because the marginal cost of each dropped to 10 minutes. Stakeholders feel informed. Surprises disappear.

Competitive intelligence on tap. The PM who knows the competitive landscape better than anyone on the team is no longer the one with the most free time. It’s the one who’s wired up the right research loops. In a space like product analytics where Mixpanel, Heap, and newer entrants are all moving fast, being the person who already synthesized what changed last week before the meeting starts is a real edge.

The goal isn’t to do more things. It’s to think at a higher level while more things get done.

08 · The Race Is Already On

One more thing from the StarCraft world worth sitting with.

When Korean pro players got access to better practice infrastructure (team houses, coaching staff, VOD analysis tools) the gap between the top players and the merely good ones didn’t close. It widened. Better tools in the hands of players who had already developed fundamentals compounded into dominance. This is not only true in the world of StarCraft, but also in real world examples: sports, companies, and more.

The same dynamic is playing out in product right now. The PMs who have strong judgment and are leaning into AI tooling are not becoming marginally more productive. They’re becoming categorically faster, operating at a level that’s difficult to keep up with using traditional approaches.

The question isn’t whether to adopt AI tools. It’s whether you’ve built the macro to take advantage of the APM increase you’re about to get.

You can give a C-rated ICCup player 300 APM, and they’ll still lose to an A-rated player at 120. The fundamentals come first. But if you’ve put in the work on the fundamentals (the judgment, the strategy, the relationships, the craft) then what’s coming is genuinely exciting.

The ceiling on individual PM leverage is moving.

You must construct additional Pylons.

Start Here: Building Your AI Craft as a PM

The gap between PMs who talk about AI and PMs who have actually wired it into their workflow is widening fast. Here’s where to start, ordered by time-to-value.

Claude.ai Projects — The fastest on-ramp. Create a Project, upload your PRD templates, customer research, and product principles. Every conversation inherits that context. Start here before anything else.

Prompt Engineering — The foundation. Learn to write system prompts and structured task prompts. Anthropic’s prompting guide (docs.anthropic.com) covers chain-of-thought, role prompting, and output formatting. One afternoon of reading compounds for years.

Claude Code — For technical PMs and anyone willing to open a terminal. Claude Code is an agentic coding environment where you describe tasks in plain English and it executes across your filesystem. Use it to build prototypes from PRDs, analyze product data CSVs, automate release notes, and explore codebases without bothering engineering. Free course at ccforpms.com.

Skills — Reusable instruction sets that encode your team’s workflows. Build a PRD skill, an interview synthesis skill, a competitive teardown skill. Each one reduces a recurring 2-hour task to a 10-minute prompt. Claude’s Skills documentation at claude.com/blog/skills-explained walks through setup.

MCP Integrations — Connect your AI workflows to the tools where work lives: Jira, Notion, GitHub, your analytics platform. Start with one integration that would save you the most weekly time. The Anthropic MCP docs (docs.anthropic.com/mcp) and Claude Desktop are the easiest entry points.

Deep Research — Claude’s built-in research mode runs multi-step web research and synthesizes findings into structured reports. Use it for competitive analysis, market sizing, and technical landscape reviews. Replaces hours of manual tab management.

The honest advice: don’t try to adopt everything at once. Pick the one workflow that costs you the most time each week and build a skill or prompt around it. That’s your hotkey. Add one per month. By the end of the year, you’ll be playing a different game than most of the field.