Early in my time at a large cloud platform, I had a product manager on my team who was struggling. She was sharp and driven, but her customer meetings kept stalling : action items didn't close, stakeholders were confused about who owned what, and sales leaders were escalating. The easy read was a coaching problem, so I sat in on her meetings expecting to hand her a few facilitation tips.
That wasn't it. She wasn't missing skills. She was missing authority : her decision rights were undefined, and the wider org didn't know to respect them. No amount of coaching fixes a structural gap. So we clarified her scope, I made her decision rights explicit to the stakeholders around her, paired her with a strong peer mentor, and we built a pre-meeting checklist for her biggest deals. Within a few months she was running executive updates on her own, and one of those accounts grew into a multimillion-dollar deal.
I wasn't thinking about a "people intelligence stack" while any of that happened. I was answering a narrower question: why is a talented person stalling, and what will actually help? The answer, almost every time, was to find the structural barrier and remove it, then get out of the way.
I led a team of about a hundred inside an org of fifteen hundred, and the part I'm proudest of isn't a number, though the manager-360 came back highest in that org two years running. It's that the approach was repeatable: assume good intent, separate the signal from the delivery, treat a mistake as a learning event and not a blame event, and remove what's in the person's way. Hiring well and keeping good people came out of that, not the other way around.
So when I went to map the HR and people-tech market : the way I did for GTM and customer success : I expected the tools to have caught up to that idea. Some have. Most haven't. And the ones that have are pointed at the wrong end of the problem.
Here's the shape of it, and it's different from the other two. GTM and CS are layered markets : a system of record, an intelligence layer, an execution layer stacked on top. HR isn't a stack. It's a lifecycle, and the AI is distributed across it unevenly. The heat is at the two transactional ends : getting people in, and answering their questions. The strategic middle : why teams perform, who's worth keeping and how, what the org should look like : is comparatively cold.
What's been built, and where it concentrates
The hiring zone, saturated
Where the capital and the startups cluster
This is the hottest area in all of HR tech, by funding and by sheer number of companies. Autonomous and AI-native recruiters : Juicebox, HeyMilo, Apriora's "Alex" : now source, screen, and run first-round interviews end to end. Talent-intelligence suites like Eightfold rank and match at scale. The incumbents consolidated hard around it: a major HRIS bought a conversational hiring agent, an enterprise suite bought a recruiting platform, a CRM absorbed an autonomous recruiter, all inside a single year.
It's genuinely impressive, and almost all of it is pre-hire. The system gets the right person in the door. What happens after they start is a different, thinner problem : the same pattern I found in GTM, where everything clustered before the first meeting.
The employee-service zone, the other saturated end
Answer the question, grounded in access
The second hot end is the internal help desk. Glean, Moveworks, and Leena AI let an employee ask a question in plain language and get an answer assembled from across every system : and, critically, grounded in what that person is permitted to see. This is the people-side equivalent of the support-automation layer in the CS market. It's fast-scaling and genuinely useful. But it resolves the ticket, not the career. Knowing where the parental-leave policy lives is not the same as knowing whether someone should be promoted, coached, or kept.
The substrate, system of record and analytics
Everything sits on a foundation built for compliance
Workday and SAP hold the enterprise record; Rippling and Deel modernized it; Visier and ChartHop turn it into analytics and org charts. The incumbents are shipping agents fast : Workday's are metered on a new credit model, which tells you they expect real consumption. But the core data was designed for payroll and compliance, for humans updating fields. An agent reads pipeline of promotions, tenure, and engagement scores as ground truth, when it actually carries all the optimism and inconsistency of how the data got entered. The most sophisticated people-analytics layer in the world is still reading from a system built for humans first.
The strategic middle, thin
Measurement everywhere, systems nowhere
Perform and engage is crowded with measurement : Lattice for structured reviews and comp, Culture Amp for engagement science, 15Five for continuous check-ins. All useful, and AI is mostly bolted on as summarization. The question my diagnostic asks : why does this team consistently outperform, and what should everyone else copy : is still answered by a human staring at a dashboard. The tools measure the middle. They don't yet run it.
Retention, scores not systems
The prediction is solved; the response is not
Flight-risk prediction is, on paper, a solved observation problem. The platforms read the leading indicators : workload, hours, internal moves, manager span, pay progression : and score who's likely to leave, with explainability layered on top. And that's exactly where the market stops. It tells you someone is at risk. It doesn't know what actually retained the last ten people like them, and it doesn't update anyone's playbook when a save works. Worse, a flight-risk score is a surveillance object by default : point it at people the wrong way and you destroy the trust faster than any attrition problem you were trying to solve. The market built a scoreboard where the real work is in the play.
Org design and workforce planning, coldest
The management operating system nobody built
How wide should spans be, where does the next head do the most good, does the structure still match the strategy : these are the highest-stakes people decisions a leader makes, and they're still made in spreadsheets and the annual restructure. Agentnoon, TeamOhana, and Gloat are starting to model org and headcount as living data rather than a slide, which is real progress. But the management operating system : the thing that helps a leader run the org as a system : is the coldest square on the board.
What's still missing, for now
When I ran the manual version of this, the hardest part was never spotting who was struggling or who might leave. It was turning what worked : the barrier I removed, the structure I clarified, the mentor pairing that landed : into something the whole org did by default. That knowledge lived in my head and my managers' heads. It propagated through coaching and tribal transfer, one person at a time, and it left when people left.
The people-tech market today has strong observation : dashboards, engagement scores, flight-risk models : and growing recommendation : agents that draft and summarize. What it doesn't have is the system that closes the loop. Not "this person is at risk" but "here's what actually retained people like them, and here's how the playbook updates for every manager." And it has to hold a harder line than the other two markets, because people data is the most trust-sensitive there is.
Three specific things are still missing:
A leader running a large org needs to know which teams are actually healthy versus which just look that way, where to spend scarce coaching attention this week, and where a resignation is quietly forming. The tools give per-employee dashboards. Nobody owns the portfolio intelligence that lets a leader run the whole org as one system.
Spans, coverage, org design, where the next head does the most good : these decisions are still made in spreadsheets and restructures, informed by experience and gut. The signal to inform them better exists. The system that turns it into a decision doesn't.
Every exit, every save, every fast-ramping manager contains information about what worked. It dies in a filed exit interview or a static risk score. The system that reads those outcomes, finds the pattern, and updates what the whole org does next quarter : automatically, not through the next enablement cycle : hasn't been built. And in people systems it has to be built with governance the current tools don't have.
Companies are moving toward this : the platforms are shipping agents, the analytics vendors are getting closer to trajectory-based signal. The "for now" in the title is deliberate. But the industry itself expects a large share of agentic projects to stall on governance and unclear value over the next couple of years, and in HR that caution is warranted. As of today, the loop is still open.
What I'm building, in what order
The HR & Org Design Diagnostic named a first set of apps. Mapping the market reshaped them : I dropped the standalone exit synthesizer and folded its signal into retention, reframed flight risk as growth, and added two the market convinced me matter. Each maps to a stage, passes the go/no-go filter, and has a data or trust gate that must hold first. The order is set by which gate is easiest to clear, not by ambition.
Let people self-serve answers grounded in what they're permitted to see. It goes first because it forces the permission model to be right on day one : and that access-aware foundation is what every later app has to inherit. This is also the least trust-fraught place to start.
Personalize the ramp by role, track milestones, flag a stall before the 90-day mark. The gate is that a structured onboarding framework has to exist first : you cannot automate a motion nobody has defined.
Personal development aligned to where the org and the team are going, not a generic course catalog. It builds on the knowledge base from the first app and sets up the real retention lever : people stay where they're growing on purpose.
Not "who's about to quit" but "how do we keep and grow this person." It reads the signals, folds in exit-interview learning and what actually retained comparable people, and points at a development or structural move : then learns whether it worked. Gated on connected data and an explicit policy on who sees what. This is the loop-closer.
The most crowded zone, so it comes last, and only with a sharp angle: calibrate against real quality-of-hire, not resume keywords. The bias gate here is not optional : biased training data produces biased outputs, and in hiring that is a legal and human problem, not just a quality one.
Beyond HR operations: leading a human-and-agent org
Everything above is about tools that help HR run its own work better: hire faster, answer tickets, score risk. But there's a larger shift underneath the whole map, and it lands on HR and org leaders whether or not they have a tool for it. Teams are no longer only human. They're becoming a mix of people and agents working side by side, and someone has to design that, manage it, and help people through it.
That's a different job from anything a recruiting agent or a help-desk bot does. It means org design that treats agents as part of the team topology, not just software: what a human owns versus what an agent owns, where the handoffs are, how wide a span can reasonably be when part of the team never sleeps. It means rethinking accountability and decision rights when an agent did the work and a person owns the outcome. It means performance and growth paths that still make sense when agents absorb the tasks that used to be how people learned the job.
And most of all it means change leadership. People don't navigate a shift like this because a policy told them to. They navigate it when someone they trust names the fear honestly, removes the barriers in the way, and treats the fumbling as learning rather than failure. That's the part of this I care about most, because it's the part least about technology and most about people. No tool on the map above does it. It's the mandate, not the software.
Why I'm building here
The HR & Org Design Diagnostic started as a way to document how I think about the talent lifecycle : the five stages, the cost of each leak, the go/no-go filter for what's worth automating. The natural next step was to build. But first I wanted to know what already exists and what's genuinely missing.
The research confirmed what I learned managing people the manual way: the market is strong at getting people in and answering their tickets, and thin exactly where it matters most : keeping good people, growing them on purpose, and designing the org around them. It builds surveillance where it should build growth. That's where I'm building, and I'm building it with the restraint people data demands.
This is the third of three : the people-side companion to the GTM Intelligence Stack and the CS Intelligence Stack. Same lens, same open loop, the part of the business that is actually made of people. For now : here's the map. Here's what's been built. Here's where I think the gaps still are.