Senior Product Manager

The word ‘soft’ was always a slight

Upfront: I'm a product manager writing about why communication and judgment matter more than technical execution. Factor that in as you like (I've just about made peace with it), but I think the argument holds regardless of who's making it.

When we talk about hard skills and soft skills, the framing isn't neutral. Hard skills are technical, measurable, teachable. Soft skills are... the other stuff. Communication. Empathy. The ability to walk into a room full of disagreeing people and leave it aligned. The work of making sure everyone understands what they're building, and why.

The implication has always been that soft skills are the easier half of the equation. Nice to have. A bonus on top of ‘real’ competence. For a long time, the job market rewarded that framing. The engineer who could write elegant code got paid more than the person who could keep a project from coming apart at the seams. The PM who could ship fast outranked the one who asked hard questions before anything got built.

Let’s unravel that conundrum. 🤓

AI is doing the ‘hard’ stuff remarkably well

The tasks that once required technical expertise are becoming accessible to anyone willing to learn the tools. A product person can generate a working prototype in an afternoon. A designer can turn a Figma frame into a functional React component. For developers, coding assistants are now writing a substantial proportion of the actual code.

Of course, that’s never been the whole story. As I wrote a few months ago, the blurring of role boundaries doesn't make deep craft skill less valuable. If anything, it raises the ceiling on what deep craft is worth. A great engineer is not less valuable because a CEO can now generate a prototype. They're more valuable, because someone has to know what to do with it once it exists.

The things AI is still terrible at

Ask AI to synthesise a set of user interview notes. It will. Ask it to identify the one assumption that could kill your product if it's wrong. It'll give you five, all plausible, at least three completely irrelevant to your specific users in your specific market. It will also be very confident about this.

The gap is in judgment.

Or take aligning a team. Getting smart, opinionated people to agree on a direction isn't a task you can delegate to a tool (I've tried 🥲). Someone has to hear what the data engineer is worried about and connect it to what sales has been hearing from customers. Someone has to notice that the head of product and the head of engineering are using the same words to mean different things, and do something about it before that misalignment ships.

That's glue work. It's not glamorous, it rarely gets named in a case study (and it’s almost never what gets you promoted), but teams that do it make better decisions than teams that don't, and the gap is widening as everything else gets faster.

Research consistently shows, as Linda Babcock wrote in 2018, that this kind of work falls disproportionately to women: more likely to volunteer for it, and more likely to be asked when they don't. The undervaluing of glue work and the undervaluing of the people most likely to do it have never been entirely separate problems, and AI making these skills newly visible doesn't automatically fix that.

The speed trap

The complexity of alignment is only getting worse and worse as speed becomes cheap and chaos reigns. Teams can now generate more artefacts, more prototypes, more options, more documents in a day than was possible in a sprint two years ago, but speed without direction is basically noise. A recent report from CircleCI shows that while engineering teams are writing dramatically more code, fewer changes are actually reaching production.

The teams struggling are the ones that have replaced slowness with pace, but skipped the conversations that slow things down productively. They've confused output with progress. You can build the wrong thing very quickly now, and it looks impressive (right up until the moment it doesn't).

What slows things down productively? A researcher who insists you talk to 5 actual users before deciding which direction to prototype. A tech lead who calls a meeting to settle what ‘done’ means before the sprint starts. A delivery manager who notices a quiet tension in the room and makes space for it instead of ploughing through the agenda. These aren't inefficiencies. They're the work that makes the rest of the work worth doing.

Glue work isn't soft. It's structural.

The term glue work was coined by Tanya Reilly in a 2019 talk at the Strange Loop conference. She described the category of work that keeps a team functioning but doesn't get you promoted: documentation, onboarding, managing the messy handoffs between teams, noticing when something is falling through the gaps and doing something about it.

What Reilly identified in engineering teams applies across every discipline. For a designer, it's the facilitation work: running the session that gets the room to a decision, rather than just delivering the screens and moving on. For a researcher, it's making findings land, not just writing them up. After all, the difference between research that changes direction and research that sits in a deck is almost always a communication skill, not a methodology one. (We’ve all got the deck. 🫡)

In the age of AI, this work isn't disappearing. It’s becoming the structure everything else depends on.

The unfair advantage hiding in plain sight

If the logic holds, you'd expect these skills to be revalued in how teams are built and hired. There are signs of it: organisations screening for judgment, job specs leading with communication over tool proficiency.

Old habits die hard though. Most organisations still optimise for what they can measure. I think the teams building the right muscle will ask different questions: not just 'can you ship fast?' but 'can you make a room smarter than when you walked in?'

The great irony of the soft skills conversation is that these skills were never soft. They were just hard to put in a job description, difficult to measure in a performance review, and easy to dismiss as table stakes rather than differentiators.

AI hasn't changed what good looks like, but it has made the consequences of ignoring it more visible, more quickly.

The teams that will thrive aren't the ones with the most access to AI tools. Those are commodities, or will be soon. The ones that will pull ahead are the ones with the best communicators, the sharpest judgment and the people who do the glue work without being asked. That combination is rare, and it's not something you can prompt your way to.

If you'd like to learn more about how Loomery can help your team build the skills that AI can't replace, get in touch.

Get the latest news and views from Loomery directly to your inbox

Stay ahead of the curve with our monthly newsletter, The Weave.

Newsletter sign up

The bottleneck isn’t the code anymore; why glue work became the hard part

May 2026

"The robots took the coding. They left us the meetings." An engineer, 2027

Get the latest news and views from Loomery directly to your inbox

Company news

Technology

Technology

Let's talk →

The bottleneck isn’t the code anymore; why glue work became the hard part

May 2026

"The robots took the coding. They left us the meetings." An engineer, 2027

Get the latest news and views from Loomery directly to your inbox

Score your team against the 8Cs

Company news

Technology

Technology

Let's talk →