Under the surface: why AI won’t fix broken HR foundations (and what will)

If you caught our last post on “AI, agents, and the human-centred future of skills,” you’ll know this much already: AI isn’t magic. It can’t replace solid human systems or fix structural rot. Today, we’re diving deeper under the surface to see why slapping “AI-powered” badges on legacy HR systems is more like turbocharging a Flintstones car. It sounds cool, but you’re still stuck pedalling.

Until recently, I hadn’t known exactly what I was articulating when I jabbered on about “broken foundations.” I’ve read, I’ve queried experts, I’ve dived headfirst into forums, LinkedIn threads, and YouTube rabbit holes, just to get a slightly clearer view of where we stand in this “Age of AI.” Spoiler: I’m not an expert. I’m a curious human trying to make sense of the hype. 

Every vendor under the sun now claims they’re “AI-powered,” “skills-enabled,” or “agent-driven.” Exciting, right?  

Only to discover, once you click “Deploy,” that nothing substantive has changed. Why? Because technology doesn’t patch structural rot. And guess what? 70% of HR tech implementations still flop at delivering on their lofty promises. 

In this post, we’ll unpack why that’s happening and what you need to do to fix the real problem “under the surface.” 

We throw the term AI around like it’s a magic wand. That’s fundamentally misleading. We’re dealing with a spectrum of capabilities, each with its own limitations: 

• AI (Artificial Intelligence): Narrow systems trained to do one thing, summarise a document, tag a photo, or filter spam. 

• ACI (Artificial Capable Intelligence): A step up, AI that can juggle multiple tasks, start to make decisions, and increase autonomy. Think Copilots, Assistant Bots, or intelligent “agents.” 

• Agentic AI: Systems that don’t just recommend; they act. They orchestrate workflows, trigger processes, or make decisions on their own. 

• AGI (Artificial General Intelligence): The mythical unicorn, human-level, self-aware, cross-domain genius. We’re not there yet, but we’re building the staircase. 

Why this matters: every so-called “agent” we see today still relies 100% on human-created data. If that data is inaccurate, the algorithm’s output will be inaccurate. If your job frameworks are spaghetti, your definitions of “skills” are ambiguous, and your systems are siloed, then no amount of flashy AI can save you. It simply amplifies the mess. 

Here’s a truth bomb: you can’t sprinkle AI over dysfunction and expect a miracle. Yet, that’s exactly what’s happening: 

• “AI-powered” learning platforms built on decades-old job taxonomies. 

• Talent marketplaces recommending internal moves that conflict with established mobility rules. 

• Inference engines generating “skills profiles” that nobody can use, because nobody agreed on what “skills” even means in the first place. 

It’s not that the AI itself is broken, it’s that the human systems underneath was never engineered for this. Our obsession with shiny dashboards and chatty bots obscures the fact that the plumbing is leaking. 

If the foundations are weak, AI won’t save you. It will accelerate chaos. 

Before you even think about agents or Copilots, get real about what’s happening under the hood. Here’s where most organisations trip: 

Outdated job architecture 

• Job families haven’t been updated in over a decade. 

• Roles are described in vague buzzwords that don’t reflect day-to-day reality. 
• No one agrees on levels or competencies, so every “AI-powered skills” module starts from zero. 

Siloed, fragmented systems 

• HRIS, LMS, performance management, talent acquisition, and payroll all live in separate kingdoms. 

• Data flows are manual, error-prone, and rarely documented. 

• Integration projects keep getting postponed because “we don’t have the budget” or “it’s too risky.” 

Undefined, fluid “skills” 

• Every leader has a different mental model of what a “skill” is. 

• Job descriptions list 20–50 “skills,” most of which are irrelevant. 

• No taxonomy. No consistent definitions. No single source of truth. 

Lack of ownership & governance 

• Who owns “skills”? HR? Line management? A random committee? 

• Data quality is an afterthought, no one’s fixing missing fields, duplicate records, or conflicting metadata. 

• No data stewards, no guidelines, no accountability. 

Misaligned KPIs & incentives 

• Your LMS vendor gets paid to log learning hours, not to ensure real behaviour change. 

• Your talent marketplace is judged on matches, not on actual internal mobility success. 

• Performance management systems reward annual reviews, not ongoing coaching or skill development. 

Every single one of these issues propels you towards faster chaos if you embrace some AI “solution” on top. So instead of chasing easy wins, “AI for everything!” let’s fix the underpinnings first. 

Here’s the disruptive truth: if you want AI to be more than a fancy brochure, you must do the hard, boring work that precedes automation. Ready? Let’s roll up our sleeves:

Rebuild your job architecture from scratch 

• Audit every role. Run focused, two- to three-week deep dives on each job family. Interview managers and front-line employees about actual responsibilities. No fluff. 

• Define core competencies. Create a lean, fit-for-purpose taxonomy of competencies aligned to strategic goals. Keep it to 15–20 core skills per function, max. 

• Level up transparently. Design clear levelling criteria (e.g., “Senior Data Analyst can influence cross-functional initiatives,” “Director of Data Strategy owns a multi-million-dollar budget”). Publish them, let people challenge them, and refine. 

Break down siloed systems 

• Map your data landscape. Draw a simple visual: HRIS ↔ LMS ↔ Performance ↔ Talent Marketplace. Note every data handoff, manual spreadsheet, and one-off API. 

• Prioritise integrations. Identify 2–3 high-impact data pipelines (e.g., HRIS → Skills Database, LMS → Talent Marketplace). Build those first. Use middleware or iPaaS tools, but get something working in 90 days. 

• Enforce data governance. Assign data stewards in each department to police data entry, fix duplicates, and enforce naming conventions. Publish a living Data Dictionary and keep it updated. 

Standardise “skills” across the org 

• Host a skills workshop. Bring together functional leaders, HRBPs, and top performers. Co-create a baseline “skills lexicon” that maps to real-world behaviors and outcomes. 

• Publish a skills taxonomy. Make it your single source of truth for all AI engines. Version it, own it, and treat it like intellectual property. 

• Embrace a “skills first” culture. Embed your taxonomy into every job posting, performance review, and learning path. If a new skill isn’t in the taxonomy, don’t add it—make it survive the workshop first. 

Establish clear ownership & governance 

• Create a data steering committee. Mix HR, IT, Finance, and a couple of line-of-business champions. Mandate quarterly reviews of data health, integration status, and roadmap priorities. 

• Define RACI for data & skills. For every data element (“Who owns employee skill ratings?” “Who audits them?”), write down who is Responsible, Accountable, Consulted, and Informed. Document it. Share it. Enforce it. 

• Implement “buy-in rituals.” Before any new AI feature goes live—like a chat-based performance coach or an automated skills profiler, require sign-off from your Data Steering Committee. No exceptions. 

Align incentives & KPIs to true transformation 

• Measure what matters. Track outcomes, not outputs. Instead of “number of courses completed,” measure “skill proficiency improvement” or “internal mobility rate.” 

• Re-think vendor contracts. Make payments contingent on demonstrable business impact, e.g., “30% increase in internal hires, measured by skills alignment,” not “we rolled out the feature.” 

• Champion continuous learning. Encourage leaders to reward managers who coach, upskill, and transfer talent internally. Embed those behaviours into your performance calibration cycles. 

When our foundations are solid, AI becomes an amplifier of human potential, not a disrupter of human pain points. Picture this: agents that autonomously scan your validated taxonomy, propose personalised career paths aligned with quarterly OKRs, and nudge managers when an employee’s skill gap matches an upcoming project. None of that is “magic” it’s simply the logical consequence of a sound foundation. 

But here’s the edgy, forward-thinking truth: nobody buys the house you’re building until they know the foundation is stable. If your board, CHRO, CFO, or CEO ask, “Why invest in AI when we barely manage our data?” you need an answer, and a blueprint. 

Stop chasing fashionable acronyms (“X-Launched Agentic AI for Learning!”). Start obsessing over the icky, boring “under-the-surface” stuff that nobody wants to talk about: 

• Dirty data 

• Outdated job frameworks 

• Undefined skills 

• Fragmented systems 

Because once those are fixed, the so-called “AI revolution” won’t be another set of empty buzzwords. It will be a genuine leap in human-centred capability, where AI agents truly amplify human skills, instead of drowning them in fast, spectacular chaos. 

We’re building that staircase to AGI. But the first few steps don’t involve cutting-edge algorithms, they involve making your human systems bulletproof. If you want AI to transform your organisation, start by fixing what’s under the surface: 

1. Rebuild your job architecture. 
2. Break down your data silos. 
3. Standardise your skills definitions. 
4. Establish clear ownership and governance. 
5. Align incentives and KPIs to outcomes. 

Only then will AI stop being a flashy veneer and become a force multiplier for your people. Because let’s be honest: without a rock-solid foundation, AI is just faster chaos. 

I’m no expert, just a curious human who will keep sharing wins, faceplants, and lessons from the bleeding edge. We’ll call out the hype, hold vendors accountable, and make sure we fix what’s broken before AI bulldozes over it. The future of skills is human-centred. If you want to ride this wave, you’d better start building a damn good foundation.