Interesting or Invisible: What Musa Tariq Taught Me About Running Programs in the AI Era.

AI has collapsed the cost of output. It hasn't collapsed the cost of taste. The next great program managers won't look like schedulers. They'll look like brand directors with a Gantt chart.

I spent the last few weeks in mentorship with Musa Tariq. Former CMO at GoFundMe, Airbnb, Ford. Brand work at Apple, Nike, Burberry before that. He now runs Science x Story with Tim Jones, a consultancy built on the idea that business strategy is brand strategy. Their client list — Stripe, Pokemon Go, Fanatics, Upwork, Kraken, Scopely — reads like a stress test of whether that premise holds up under weight.

I went in to sharpen my positioning. I came out rewiring how I think about program management.

One line of Musa's kept coming back, in different forms, across our sessions: brands die from being generic, not from being wrong. What surprised me was how directly it applies to the operational work I do every day. The AI era has quietly turned every program manager into a brand custodian. Nobody signed up for it. It happened anyway.

The Abundance Problem Nobody's Naming.

Most of the AI-and-PM conversation is still about speed. AI drafts the status update faster. AI spits out roadmap options faster. AI generates the retro notes. Fine. That framing isn't wrong, but it's missing the part that's becoming operationally dangerous.

In January, researchers from BetterUp Labs and Stanford's Social Media Lab published a follow-up in HBR on what they'd named workslop: AI-generated work that looks polished but lacks the substance to move anything forward. The numbers are ugly. 41% of US desk workers said they'd received workslop in the past month. Each incident cost the recipient about two hours of rework. For a 10,000-person company, that's roughly $9 million a year in lost productivity.

Merriam-Webster named "slop" its 2025 Word of the Year. The MIT Media Lab reported 95% of corporate AI pilots show no measurable return. Read these numbers as a consumer problem and you miss the point.

The same thing is happening inside your organization. Decks, briefs, retro notes, project charters, quarterly reviews. All of it rides the same rails as AI-generated marketing. The baseline quality of anything machine-assisted is competent-and-forgettable. When every program ships at that baseline, programs blur together. Priorities blur. Teams lose the thread. Nobody can quite say why the work feels directionless, even though every artifact looks professional.

Musa's version of this, roughly paraphrased from our sessions: the cost of producing something that looks right has gone to zero. The cost of producing something that sounds like you has never been higher.

What Musa's POV Does to the PM Role.

Musa's lens on brand is that it isn't a visual system or a tone guide. It's a set of decisions about who the business is for, what it stands for, and what it refuses to do. The Science x Story method starts a step earlier than most brand work: figure out what the business model requires of the brand before you decide what the brand stands for. That sequencing matters. It forces the brand to be load-bearing instead of decorative.

Apply that to program management and the role bends in a useful direction.

A program is also a set of decisions about who the work is for, what it stands for, and what it refuses to do. A PM who only orchestrates — who runs their role as scheduling, status, and cross-functional traffic — is running a brand-less program. It can be efficient. It can hit dates. It can't be recognizable. And when AI is producing plausible versions of everything, the recognizable work is the only work that compounds.

This maps onto what HBR has been tracking all year. Their February piece on how workers develop good judgment in the AI era lays out the trade: AI handles the messy repetitive tasks that used to build judgment in junior people. Organizations now risk producing managers who've never done the underlying work. Their April follow-up, decision-making by consensus doesn't work in the AI era, is blunter. When AI can generate an infinite number of defensible options, consensus becomes a path to the most generic one. Someone has to have a point of view. Someone has to have taste.

That someone, increasingly, is the PM.

Taste as an Operational Artifact.

The phrase "taste is the new moat" has been in the air for a while. Designative wrote it up well in February. Most of the discourse lives in design and founder circles. The translation into program operations hasn't happened yet. It's overdue.

Here's what taste-as-operational-artifact looks like in practice.

A program has a voice, whether you author it or not. Your standup format, the shape of your weekly notes, how you frame a risk, what gets escalated and what gets absorbed, which meetings exist and which ones you kill. Those are editorial decisions. They tell the team and the exec audience what kind of program this is. Run them on AI defaults and you get an AI-default program: polite, comprehensive, forgettable. The team will disengage long before they can say why.

A program has a refusal set. This is the Musa-brand equivalent I've been running with teams lately. What won't this program say yes to, even when asked? Not in a precious way. In a strategic way. The refusal set is what makes the yeses mean something. Without it, every request lands on the backlog with the same weight, and AI will happily help you generate an increasingly elaborate backlog of things that don't matter.

A program has an exec experience. Every artifact the program ships — every Loom, every Notion page, every Slack update — is a touchpoint. In a pre-AI world, that framing would've sounded precious. Now, when your VP is receiving the same AI-polished updates from every program in the org, the programs with intentional voice are the ones that get remembered, funded, and expanded. The rest quietly compete for attention they won't win.

The throughline: in a world of infinite plausible outputs, the PM's real deliverable is editorial judgment. Which of the ten agent-generated roadmap variants belongs in the world. Which of the fifty AI-summarized risks actually matters. Which auto-generated meeting notes captured the decision correctly, and which one got it subtly wrong in a way that'll bite the team in three weeks. That work isn't a nice-to-have. It's the work.

The New PM Stack: Process + Taste + Prompt.

Old PM stack was roadmap, rituals, risk. The emerging one keeps that and adds two layers on top: taste and prompt.

The taste layer is what we've been talking about. It's the editorial function. Musa-type questions baked into program cadence: Is this interesting? Does this sound like us? Would anyone miss it if we didn't ship it? Those belong on the standing agenda next to budget and scope. They aren't soft questions. They're load-bearing.

The prompt layer is the literal one. The PM who hand-writes every status update is about to get outcompeted by the PM who built a well-tuned prompt system that drafts them, then exercises taste on the output. That's not lazy. It's leveraged. HBR's research on what the best AI users do differently is clear on what separates the top performers: it's the specificity of the request, and the willingness to treat AI as a reasoning partner instead of an autocomplete. PMs who learn this become force multipliers. PMs who don't become middleware humans, routing between AI outputs and human decisions. That role will get automated by the same agents they're managing.

The prompt layer is useless without the taste layer above it. A well-prompted system that produces beautifully generic content is worse than a rough human draft with a real point of view. HBR's February piece AI doesn't reduce work, it intensifies it names the trap cleanly: AI offloads execution but loads up the judgment, review, and verification work. No taste, no way to do that work well. You just ship faster slop.

The Real Question for the Next Five Years.

The obvious version of the AI-and-PM conversation is about agents taking over tasks. That's real. Agents Today tracks the reshuffling well: the polarization of PM roles into strategic work on one end and agent-orchestration work on the other, with the middle disappearing.

The more interesting version is the one Musa's work points at. When execution gets cheap, the scarce resource is conviction. A point of view. The willingness to be specific when the machine will happily be general for you.

Brands die from being generic. So do programs. So, eventually, do the careers of program managers who never build the taste to tell a competent output from a memorable one.

The world is filling up with plausible work. Your program, the one with your name on it, is either part of the noise or part of the signal. The tools won't decide that. The decision is editorial. The decision is taste. The decision is, as Musa would put it, whether you're going to be interesting, or whether you're going to be invisible.

I know which side I want to be on. I suspect you do too.

Further Reading: Science x Story · Former Airbnb and 72andSunny Execs Team Up to Bridge the Brand-Business Divide (Adweek) · Why People Create AI "Workslop" — And How to Stop It (HBR) · Why AI 'workslop' is bad for employees and costs businesses millions (IT Pro) · How Employers Can Get Smarter About AI Use in 2026 (Fisher Phillips) · How Do Workers Develop Good Judgment in the AI Era? (HBR) · Decision-Making by Consensus Doesn't Work in the AI Era (HBR) · What the Best AI Users Do Differently (HBR) · AI Doesn't Reduce Work — It Intensifies It (HBR) · Taste Is the New Bottleneck (Designative) · The Great Reshuffling: How AI Is Polarizing Product Management Roles (Agents Today)