Why Job Architecture is Failing Us

Blame AI all you want, but this isn’t a new thing

Jobs change. Roles change. Skills change. Work changes. Life changes. That’s sorta how the world works.

Job architecture in particular is going through a reckoning, though. AI is accelerating change to the point where comp teams can’t figure out how to price a job.

And it’s easy to blame AI for all of this. I know I love to blame AI like I blame my hair for not covering my head. But I think this is actually an evergreen problem that has only been accelerated by AI.

ADP Research and Stanford's Digital Economy Lab tracked about 7,000 IT workers from 2019 to 2025 using real payroll data. The researchers looked at 5 million job postings and 9 million payroll records across more than 600 employers. They then mapped the work into 25 standard task categories and ran the wage data against each one.

They also did some controls to figure out how much of it wasn’t due to age, gender, or company size.

Their key finding? The work changed but not in the way they expected.

PwC's 2026 Global AI Jobs Barometer, released around the same time as the ADP-Stanford research, is the perfect companion for explaining what changed. They found that roles split across two broad categories.

Professionalized roles use AI to sharpen expertise that’s already there, like a radiologist or a recruiter. Those roles are growing at twice the market rate and their pay is climbing 42% faster.

Democratized roles use AI to let someone without the expertise do the job anyway, such as an IT service manager, a medical secretary. Those roles are growing slower, and so is the pay attached to them.

Those changes work their way down to the entry-level, too. Workers with highly democratized roles are struggling but so too are the entry-level folks in professionalized roles. The latter are expected to show up with senior knowledge at a discount rate.

Not awesome.

But behind all of this is a truth: We’ve always lagged on job architecture. Things change, we work, and the reference documents get updated… eventually. AI hasn’t really transformed a thing that wasn’t already happening. It just added fuel to the fire.

There is a way forward for those who are interested. I wrote all about it on Reworked last week. I love diving into this stuff when given the opportunity.

What else is happening this week?

• America’s Jobs Scare Recedes. August came in at 162,000 jobs, more than triple what economists expected, and Courtenay Brown and Neil Irwin read it as the summer scare being seasonal noise. I’m not sure that’s right, since a three-month average of 71,000, up from 20,000, is a recovery from bad and not much more.

• Job Seekers Are Hiding Invisible AI Prompts Inside Their Resumes. The old white-text-on-white-background trick is back, this time as instructions to the AI screener. Duke’s Neil Gong found it in about 1% of 200,000 resumes on hireEZ, up sevenfold since 2024.

• AI Interviews Rolling Out for Some Federal Government Hires, Sources Say. The federal government is starting AI-run structured interviews through CodeSignal for the Tech Force program. Scott Kupor‘s OPM has already encouraged agencies to bring AI into hiring with a human still in the loop, and officials expect more of it. Good luck?

• Salesforce Leads $166M Investment in HiBob. $166 million led by Salesforce at a $3.2 billion valuation, and Ronni Zehavi‘s pitch is the bigger part: HiBob becomes the source of truth on roles, skills, and reporting lines for whatever AI agent happens to be asking.

• When Is a Protest in the Workplace Not a “Workplace” Protest? NLRB Backs Microsoft’s Termination of Employees Who Organized Gaza Demonstration. The Board’s Division of Advice found the on-campus protest wasn’t tied closely enough to working conditions to be protected, even with a months-old petition about Azure contracts behind it. A workplace demand now has to be the actual objective of a protest rather than an afterthought, which narrows the definition.

• The Application Is Not a Strategy. Brian Fink puts numbers on it: applicant-to-interview rates have fallen from around 15% in 2016 to 3% in 2024, and people who send more than 81 applications land offers less often than people who send fewer. The application was never the strategy, it’s just harder to pretend otherwise now.

• An Alien Mind: OpenAI’s Chief Scientist Jakub Pachocki Concerned About Astra, Should We Be?. Probably. Josh Bersin reads the post on OpenAI’s new Astra model as honest about alignment definitions that don’t account for values, and asks how you judge a model’s values when you don’t know what it was trained on.

• “No White After Labor Day” and Other Authorless Rules. Loved this one. Robin Schooling on the rules nobody explains: the $25 expense threshold from an incident in 2011, the no-meetings-before-9:30 rule that outlived the executive who wasn’t a morning person, all of them still standing because asking costs more than complying.

• Why People Resist Technological Innovation. Senior leaders have 20-plus years of watching implementations fail, middle managers fear anything added to a job that already runs long, and frontline employees assume change means worse work or none. Steve Hunt would have you answer each of those before the rollout, which is the right set of questions to be asking.

• Mediabistro Weekly Drop: The New Drawn Edition. I love what Matt Charney is doing at Mediabistro, and this one gets at what happens when accountants hold final cut over the people doing the work.

• Candidates Don’t Trust AI in Hiring. That’s Our Problem to Solve.. Matt Alder is an optimist about AI in hiring and still leads with the data against him: ThriveMap found 49% of UK job seekers think it’s unfair for employers to use AI in recruiting and 41% want no AI in the decision at all.

• Talent Management May Be Dead. But HiBob Just Moved to the Front of Whatever It’s Called Next. Steve Smith was drafting an obituary for the talent management category when the Salesforce news landed. His thesis is that the software is toast and the data underneath it is the bigger opportunity, and yep, this is big.

• Research Note: Engagedly/Energage. Srikant Chellappa sold Engagedly to Energage, which runs Top Workplaces, and asked for an analyst to look hard at the deal. Strong note from John Sumser, who obliged with the question UKG ran into after buying Great Place To Work: do I have to buy your software to win your award?

• The AI Boundary Most Professionals Haven’t Drawn, but Should. Danielle Farage surveyed 1,000 Gen Z professionals and found 67% deliberately protect certain skills from AI. Mervyn Dinnen writes up her three-zone test for any task (delegate, collaborate, or protect), and the protect zone is the missing one for most.

• You Can’t Spell Hormones Without HR. Retirement gets a webinar, a calculator, three reminder emails, and a rep from Fidelity, while a hormonal transition that will hit a large share of any workforce gets silence. I laughed, then nodded, at Madison Butler‘s take.

• Hold Still While I Talent Advise You. TA leaders keep telling recruiters to become talent advisors, and John Vlastelica‘s point is that hiring managers often have no idea what to do with one. You can’t order off a menu you’ve never seen.

• A Way Too Early HR Tech Preview. Never too early, and Mike Wood expects agents in every booth this fall, with Madeline Laurano‘s judging showing only 12% of companies reporting agentic AI use and nearly every product submitted claiming it.

• Making Claude and Perplexity Part of My Team. Good to see Jonathan Goodman writing out in the wild. On the first engagement under his new business he gave Claude and Perplexity four named roles with explicit things each wasn’t allowed to do, then fired a fifth when its summaries flattened three executives’ distinct positions on one word into a single view.

• Cathy Engelbert Is Retiring as WNBA Commissioner. Leela Srinivasan on the WNBA commissioner stepping down at year end after seven years that, on valuations, media rights, and pay, put the league ahead of every other women’s league in North America. A change in leadership is a change in opportunity, and what got the W here won’t be what gets it to the next level.

• How to Build an AI Development Team That Can Deliver at Scale. My piece for X-Team’s Keep Moving Forward magazine, and you should read it anyway. Governance and domain expertise came up early and often, alongside architecture, when I asked five people who build these teams.

• Are We Watching the Economics of Work Start to Shift?. A good August recap that looks across hiring, layoffs, pay, benefits, and who’s participating, and asks whether the separate headlines add up to something bigger, from Mel Plett and Francesca Ranieri.

• Superpowered Launches in October. A first book on what AI does to individuals, teams, and organizations, written with Josh Bersin and out next month. Congrats to Kathi Enderes.

• Introducing Inspiring Place to Work. Matt Manners is turning the award into an identity: from 2027 every organization that meets the standard can call itself an Inspiring Place to Work, and 2026 winners get the badge now. That’s a big move for a program that has been running since 2015.

• The Meg & Amy Show Is 31 Followers Short of 1,000. Keep listening to Meg Bear and Amy Wilson, whose premise is that sensemaking is a team sport and collective intelligence beats individual expertise in times of change.

• The Global Talent Strategy and Intelligence Conference, Amsterdam, September 28 to 30. Jess Von Bank is emceeing Intelligence Group’s gathering at the H’ART Museum, with talent leaders from Philips, Booking.com, NATO, and Microsoft on the program. And yes Jess, you can have the pumpkins out now, as long as they aren’t real ones.

• The HR Morning Show Hits Episode 100. Evan White marks episode 100 of the HR Morning Show, featuring a baseline standard for ending candidate ghosting, a former USWNT player on high-performing teams, and the producer’s last turn in the chair. Congrats to the whole crew.

Have a great rest of your week!