Building Teams That Stay Ahead

AI Talent Strategy and Skills Development

The Lawyer Who Saved Her Firm, and the One Who Didn't

In early 2023, two mid-sized law firms in Chicago faced the same pressure: clients were asking why their legal bills were so high when AI tools could apparently draft contracts in minutes. At Morrison & Foerster's Chicago office, managing partner Laura Sheridan didn't panic. She ran a 90-day internal experiment. Every associate was given access to Harvey AI, a legal-specific AI tool built on GPT-4, and asked to use it on real client work under supervision. Associates who embraced it cut first-draft contract time by 60%. Clients noticed. Billing rates held. The firm positioned itself as the one that used AI to do smarter work, not cheaper work. Sheridan didn't hire a single new technologist. She developed the talent she already had.

Twenty miles away, a competing firm took the opposite approach. Leadership decided AI was a technology problem, not a people problem. They bought an enterprise license for a general AI writing tool, sent one email announcing it was available, and waited for adoption to happen organically. Twelve months later, fewer than 15% of staff had used it more than twice. Senior partners complained it produced 'generic' output. Associates said they didn't trust it on client-facing work. The tool sat largely unused while the firm continued losing pitches to competitors who could turn proposals around faster. The software was identical. The strategy was completely different.

This is the central tension in AI talent strategy: most organizations treat AI adoption as a procurement decision when it's actually a learning and development decision. The tools are widely available. ChatGPT Plus costs $20 a month. Microsoft Copilot is bundled into Microsoft 365 licenses that millions of companies already pay for. The differentiator is never the tool, it's whether your people know how to use it well, trust it enough to rely on it, and have been given the time and permission to build those skills. That's a management challenge, not a software challenge.

The Skill Gap Is Real. But It's Not What You Think

The actual skill gap in most organizations isn't technical. It's what researchers at MIT Sloan call 'AI fluency', the ability to work alongside AI tools effectively, evaluate their outputs critically, and know when human judgment is irreplaceable. A 2023 study by Microsoft and LinkedIn found that 82% of business leaders said they'd prefer to hire someone with AI skills over someone without, even if that person had less direct experience in the role. But only 14% of employees said their employer had provided any formal AI training. That gap, between what leaders want and what they're actually building, is where most companies are losing ground.

The story of IBM also illustrates a principle that runs through every successful AI talent strategy: you have to be specific about which skills you're building, for which roles, and for what purpose. 'Everyone should learn AI' is about as useful as 'everyone should learn Excel.' True, but it tells you nothing about what to teach a regional sales manager versus a junior HR coordinator versus a marketing director. The firms and teams that get this right build role-specific AI skill maps. They identify the three to five tasks in each job function where AI tools create the most leverage, then train specifically for those tasks. Everything else follows from that clarity.

How Unilever Rebuilt Its HR Skills Pipeline Around AI

Unilever's HR transformation is one of the most documented examples of AI skills development at scale, and it's instructive precisely because it happened in a function that most people wouldn't associate with cutting-edge technology. Starting around 2019 and accelerating through 2022 and 2023, Unilever began using AI tools throughout its talent acquisition process: automated video interview analyzis, AI-assisted resume screening, and predictive tools to identify internal candidates for promotion. The critical decision Unilever made was that its HR professionals needed to understand these tools well enough to challenge them. Every recruiter went through training not just on how to use the AI tools, but on what they couldn't do, what biases to watch for, when to override the system, and how to explain decisions to candidates.

The result was a team that didn't just use AI, they became more effective advocates for fair hiring because they understood the tool's limitations. Unilever reported reducing hiring time by 75% for entry-level roles while also increasing the diversity of candidates who made it to final rounds. Those outcomes don't happen by accident. They happen when an organization invests in building genuine AI fluency in its people, not just access to AI software. The Unilever HR team's AI skill wasn't technical, it was the professional judgment to use AI as a starting point and apply human expertise to the decisions that mattered most.

What Good AI Skills Development Looks Like Across Functions

Click to enlarge

The Sales Manager Who Turned a Skeptical Team Around

Derek Paulsen manages a regional sales team of 11 people at a mid-sized logistics company in Atlanta. In early 2024, his company rolled out Salesforce Einstein, an AI layer within their existing CRM that could summarize account histories, suggest next best actions, and draft follow-up emails. Paulsen's team was skeptical. Several reps had been in the industry for over a decade and felt the AI outputs were 'too generic' for the relationship-driven business they were in. Two reps quietly stopped using it after the first week. Paulsen didn't mandate usage or escalate to leadership. Instead, he spent 30 minutes with each rep individually, looking at their specific accounts and asking the AI to generate outputs for real situations they were dealing with right now.

Within three weeks, eight of his eleven reps were using Einstein daily. The shift happened because Paulsen taught his team a single skill: how to add specific context to make AI output useful. Instead of asking the tool to 'write a follow-up email,' his reps learned to feed it the client's name, their last conversation topic, the specific concern raised, and the outcome they were trying to achieve. The output went from generic to genuinely useful. By Q3 2024, his team's average proposal turnaround time had dropped from four days to one and a half. Paulsen's lesson: AI skills development isn't a training event. It's a coaching conversation, done in context, with real work.

Building an AI Skills Strategy That Actually Sticks

Most AI training programs fail for one of three reasons: they're too broad, they're too theoretical, or they happen once and are never reinforced. A McKinsey report on workforce reskilling found that programs with the highest long-term adoption shared three characteristics, they were tied to specific job outcomes, they included manager involvement, and they were delivered in short, repeated sessions rather than single day-long workshops. That pattern holds up across industries. A one-hour monthly session where your team practices AI on a real task from their workflow will outperform a half-day AI bootcamp every time. The bootcamp feels more substantial. The monthly session actually changes behavior.

The second critical element is creating what some L&D professionals call a 'safe to fail' environment around AI experimentation. Many professionals, especially experienced ones, resist AI tools because they don't want to produce inferior work. They'd rather take four hours to write something themselves than risk sending a client a mediocre AI draft. The way to overcome this is to make AI practice a team activity, not a solo performance. When a marketing manager shares an AI-generated draft in a team meeting and the group improves it together, the social pressure flips: trying AI becomes the norm, and not trying it becomes the outlier behavior. This is a culture intervention, not a technology one.

The third element is tracking the right metrics. Many organizations measure AI adoption by license usage, how many people logged into the tool. That's the wrong metric. What you want to track is time-to-output on specific tasks. How long does it take your team to produce a first draft of a client proposal? How long does it take to summarize a 60-minute meeting into a one-page action list? How many rounds of revision does a job description typically need before it's approved? Establish your baselines now, run AI-assisted workflows for 60 days, and measure again. Those numbers will tell you whether your AI skills development is working, and they'll give your team a concrete reason to keep going.

Click to enlarge

Key Principles from Part 1

1. AI adoption is a learning and development challenge, not a procurement challenge. Buying licenses doesn't build skills.
2. The real skill gap is AI fluency, knowing how to use tools well, evaluate output critically, and apply human judgment at the right moments, not technical expertise.
3. Role-specific AI skill maps outperform generic 'everyone learn AI' mandates. Identify the three to five tasks per role where AI creates the most value and train specifically for those.
4. Skills development works best when it's tied to real tasks in your team's actual workflow, not hypothetical exercises or one-off workshops.
5. Manager involvement is the single biggest predictor of whether AI skills training sticks. Teams follow what their managers model and reinforce.
6. Track time-to-output on specific tasks, not license usage. That's the metric that tells you whether skills are actually developing.
7. Creating a team-level 'safe to fail' environment, where AI experimentation is practiced together, not judged individually, accelerates adoption faster than any software rollout.

Building the Bridge: From AI Awareness to AI Fluency

In 2023, Unilever's HR leadership faced a familiar problem with an unfamiliar edge. They had rolled out Microsoft Copilot to roughly 2,000 employees across marketing, procurement, and HR. Adoption numbers looked fine on paper, licenses activated, logins recorded. But six months in, a quiet audit revealed something uncomfortable: most employees were using Copilot for exactly one thing. They were asking it to summarize emails. A tool capable of drafting supplier negotiation briefs, analyzing engagement survey data, and building first-draft job architectures was being used as a slightly smarter inbox filter. The investment was live. The transformation was not.

Unilever's internal learning team dug into why. The answer wasn't resistance. It wasn't technical failure. It was a skills gap that sat just below the surface, employees didn't know what they didn't know. They'd received a 90-minute onboarding session, a PDF of use cases, and a link to Microsoft's help center. That was it. Without structured practice, without role-specific examples, without someone showing a procurement manager exactly how to use Copilot to benchmark supplier pricing or draft RFP language, the tool stayed shallow. Awareness had been achieved. Fluency had not.

Unilever's response became a case study in deliberate skill-building. They created what they called 'AI Champions', one per business unit, selected not for technical background but for curiosity and peer credibility. Each champion received deeper training, then ran monthly 'AI in Practice' sessions tailored to their team's actual work. Procurement champions showed procurement scenarios. Marketing champions showed campaign workflows. Within four months, active feature usage per user tripled. The lesson wasn't about the technology. It was about translating generic capability into specific, role-relevant skill.

What AI Fluency Actually Looks Like in Practice

Think of AI fluency the same way you'd think about business writing fluency. Everyone in your company can type. Very few can write a compelling executive summary. The difference isn't access to a keyboard, it's knowing structure, audience, purpose, and tone. AI fluency works identically. Most professionals can type a question into ChatGPT. Far fewer know how to frame a prompt that produces something genuinely useful for their specific job. The skill isn't technical. It's communicative, knowing how to brief an AI the way you'd brief a sharp but uninformed colleague.

AI fluency for non-technical professionals has three distinct layers. The first is tool familiarity, knowing which tool to open for which job, understanding what it can and can't do, and being able to navigate its interface confidently. The second is prompt craft, the ability to give an AI enough context, constraints, and direction to get output that's actually useful without heavy editing. The third, and most valuable, is workflow integration, knowing where AI fits in your existing process, where it doesn't, and how to combine AI output with your own judgment to produce better results faster. Most training programs cover only the first layer.

The organizations building durable AI capability are investing in all three layers, sequentially and repeatedly. They're not running one-off workshops. They're embedding AI practice into the rhythm of work, monthly team challenges, peer sharing sessions, manager-led debriefs where teams discuss what worked and what didn't. Fluency compounds. A marketing manager who practices AI-assisted brief writing for three months develops instincts that a one-day bootcamp graduate simply doesn't have. The organizations winning at AI talent strategy understand that skill development is a calendar commitment, not a training event.

Click to enlarge

A Retail Giant's Different Problem: Skills at Scale

Walmart's AI upskilling challenge looks nothing like Unilever's. Where Unilever was working with thousands of knowledge workers in office roles, Walmart needed to build AI literacy across a workforce of 1.6 million U.S. associates, many in stores, many without regular computer access, many working part-time across wildly different roles. In 2023, Walmart announced a $1 billion investment in employee training and technology, with AI literacy explicitly included. Their approach didn't try to turn store associates into prompt engineers. It focused on something more targeted: helping frontline managers understand how AI-generated inventory forecasts and scheduling recommendations worked, so they could act on them confidently rather than override them out of habit.

Walmart's training was built around decision points, not tool features. A store manager doesn't need to understand how a demand forecasting model works mathematically. They need to know: when should I trust this recommendation, when should I question it, and what do I do when it seems wrong? That framing. AI as a decision support tool, not an authority, changed the training entirely. Sessions focused on scenarios: here's what the AI suggested for staffing this weekend, here's the data it used, here's when you'd override it and why. The result was managers who were neither blindly deferential nor reflexively skeptical. They were informed collaborators with their own tools.

Comparing AI Skill Development Approaches

Click to enlarge

The Teacher Who Rewrote Her Entire Workflow in 60 Days

Tara Chen teaches 10th grade English at a public high school in Austin, Texas. She's not a technology enthusiast. She describes herself as 'someone who still prefers paper gradebooks.' In January 2024, her district rolled out a Google Workspace for Education pilot that included Gemini AI. Tara attended a mandatory 2-hour training, left skeptical, and mostly ignored the tool for three weeks. Then a colleague showed her something specific: how to use Gemini to generate differentiated reading comprehension questions at three different complexity levels from a single text passage, in under two minutes. That one demonstration changed everything.

Within 60 days, Tara had rebuilt significant chunks of her workflow around AI assistance. She used Gemini to draft parent communication emails (then edited them for her voice), generate first drafts of rubrics, create discussion question sets for novels, and summarize student essay feedback patterns across a class set. She estimates she reclaimed eight to ten hours per week. Critically, she didn't become a 'tech person.' She became a skilled user of one tool, applied to work she already knew deeply. Her subject matter expertise remained the judgment layer. Gemini handled the volume work. Tara's story illustrates something important: the barrier to AI fluency for non-technical professionals is almost never technical aptitude. It's almost always the absence of one specific, relevant demonstration.

Designing an AI Skills Program That Actually Sticks

The difference between AI training programs that produce lasting behavior change and those that produce politely completed modules comes down to three design choices. First, specificity beats generality every time. A training session called 'Introduction to AI Tools' produces awareness. A session called 'Using Copilot to Write Better Client Proposals in Under 30 Minutes' produces behavior change. The more precisely you connect training to the actual work your people do every day, the more likely they are to practice, improve, and integrate. Generic AI literacy content has its place, but it cannot be the whole program.

Second, practice must be built into the program design, not left to individual motivation. Adults don't develop professional skills by watching videos and hoping to find time to practice later. The most effective AI skill programs embed practice directly into training sessions, participants work on real tasks from their actual jobs using AI tools during the session itself, not on hypothetical exercises. A sales manager practices drafting a real account plan. An HR director practices building a real job description. The output they create during training is something they can actually use. That immediate utility is what triggers continued use after the training ends.

Third, the organizational environment must signal that AI skill development is valued, not just permitted. When managers openly discuss how they use AI in their own work, when team meetings occasionally include a two-minute 'AI tip of the week,' when performance conversations acknowledge AI-assisted productivity gains, the message is clear: this is a real professional skill, not a side experiment. Conversely, when leadership never mentions AI, when managers seem uninterested in their team's tool usage, and when the only AI conversations happen in IT channels, the implicit message is that it's optional. Culture moves in the direction that leadership visibly invests attention.

Key Principles from This Section

1. Access without fluency produces shallow adoption, licenses and logins are not evidence of skill development.
2. AI fluency has three layers: tool familiarity, prompt craft, and workflow integration. Most programs address only the first.
3. Role-specific training outperforms generic AI literacy content in every measurable outcome, practice rate, retention, and behavior change.
4. Internal AI Champions, selected for peer credibility, not technical background, are more effective than external vendors for sustained team-level adoption.
5. One specific, relevant demonstration converts skeptics faster than any amount of overview content.
6. Skill development programs that embed practice during the training session produce better outcomes than programs that rely on post-training self-study.
7. Leadership behavior, not policy documents, determines whether AI skill development is treated as a real professional priority or an optional experiment.
8. At scale, AI training must focus on decision support literacy: when to trust AI recommendations, when to question them, and how to act on them with informed judgment.

What to Take Forward

The organizations pulling ahead on AI capability right now are not necessarily the ones with the most sophisticated tools or the biggest budgets. They're the ones that treat AI skill development as a continuous management practice, something that lives in team meetings, in one-on-ones, in how work gets reviewed and discussed. The gap between AI awareness and AI fluency is real, it's measurable, and it's closeable. But it closes through repeated, role-specific, practice-based learning, not through a single training event, however well-designed. Your team's AI fluency is a direct reflection of how consistently you invest attention in building it.

Building Skills That Stick: Making AI Learning Part of the Job

In 2023, Walmart made a decision that most large retailers hadn't dared to attempt: they embedded AI skill-building directly into shift work. Instead of sending managers to off-site training programs, they deployed AI-assisted learning tools inside the daily workflow, so a store manager practicing a difficult employee conversation could get coaching from an AI simulation between inventory checks. The result wasn't just higher completion rates on training modules. Managers reported actually retaining what they learned because it was connected to a real problem they faced that week. Walmart's approach exposed a tension most organizations are still wrestling with: formal AI training programs get scheduled, attended, and forgotten. Learning that's woven into real work gets used.

The principle Walmart demonstrated is deceptively simple: AI skills don't develop in classrooms, they develop in context. When a marketing manager uses ChatGPT to draft a campaign brief and then edits the output, she's building prompt judgment. When an HR director uses Copilot to summarize 40 interview notes before a hiring decision, he's learning what AI gets right and where it drifts. These micro-moments of use, reflection, and correction are the actual engine of AI competency. Organizations that wait for a formal 'AI training program' to launch before expecting staff to engage with these tools are creating a skills gap by design.

The extracted principle from Walmart's model is this: the best AI talent strategy isn't a curriculum, it's a culture of low-stakes experimentation with real work. That means giving people permission to try, tools to access, and enough psychological safety to share what didn't work. It means treating a bad AI output as a learning moment, not a reason to distrust the technology entirely. Organizations that build this culture find that AI competency distributes itself naturally across teams, rather than concentrating in a single 'AI champion' who becomes a bottleneck.

When Teams Learn Together, Skills Scale Faster

Etsy's product and content teams ran an informal experiment in 2023 that became a model worth studying. A small group of content strategists started sharing weekly 'prompt diaries', short Slack posts documenting what AI prompt they used, what the output looked like, and what they changed before using it. No formal program. No manager mandate. Within eight weeks, the practice had spread to three other departments. Junior writers were learning from senior strategists. Strategists were borrowing prompt structures from the SEO team. The shared vocabulary around AI quality, 'that output was too generic,' 'I had to add more context about our buyer', started shaping how the whole team talked about their work.

What Etsy's teams discovered is that peer learning is the most efficient AI upskilling mechanism available to most organizations right now. Formal AI certifications take weeks. Vendor training sessions cover features, not judgment. But a colleague who shows you the exact prompt she used to cut her weekly report from three hours to 45 minutes? That you remember. That you try on Tuesday. The social dimension of skill-building, watching someone better than you work, asking why they made that choice, copying and adapting, is exactly what accelerates AI fluency across a team.

The mechanism doesn't have to be Slack. It can be a standing 15-minute 'AI moment' at the start of a team meeting, where one person shares something they tried. It can be a shared document where salespeople log the prompts that worked for proposal drafting. It can be a manager who asks, at the end of a project debrief, 'did anyone use AI on this, and what did you learn?' The format matters less than the habit. Teams that create structured moments for sharing AI experiences build collective intelligence that no training program can replicate.

Different Roles, Different Skill Gaps

Unilever's HR function ran one of the more documented AI upskilling initiatives in the consumer goods sector, targeting recruiters specifically. The challenge wasn't technical, their recruiters were willing to use AI tools. The challenge was judgment: knowing when to trust an AI-generated candidate summary and when to dig deeper, recognizing when a job description drafted by AI was subtly biased, understanding that AI-ranked shortlists reflected historical hiring patterns rather than future potential. These aren't software skills. They're critical thinking skills applied to AI outputs, and they required a completely different training approach than 'here's how to use the tool.'

Unilever's solution was scenario-based calibration: recruiters would review AI-generated outputs alongside human-generated ones, without knowing which was which, and then discuss the differences as a group. This built the evaluative muscle that pure tool training never develops. The lesson for any organization is that AI skill gaps aren't uniform. A salesperson needs different AI fluency than a finance analyzt. A teacher using AI to differentiate lesson plans needs different skills than a consultant using it to synthesize research. Skills development that treats everyone the same produces mediocre results across the board.

From Individual Skill to Organizational Capability

The shift from 'some people on my team use AI' to 'our team operates with AI fluency' is a leadership question, not a technology question. It requires managers to model AI use openly, to say in a meeting, 'I used Claude to draft the options framework I'm showing you, and here's what I changed.' It requires organizations to stop treating AI tool access as a perk for certain roles and start treating it as standard equipment, the way they treat email or a laptop. And it requires a recalibration of what 'good work' looks like, not dismissing AI-assisted output as less legitimate, but evaluating the judgment, editing, and contextual intelligence the person brought to shaping that output.

Organizations that have made this shift, even partially, report a compounding effect. When AI fluency is normal, people share discoveries faster. When sharing is normal, the average skill level rises. When skill rises, people find more sophisticated uses. Within 12–18 months, teams that started with basic prompt experimentation often develop proprietary workflows, specific prompt sequences, custom instructions, and AI-assisted processes, that become genuine competitive advantages. That compounding only starts when leadership treats AI skill-building as a strategic priority rather than an optional personal development activity.

The practical starting point is modest. You don't need a formal AI training program to start. You need three things: access (make sure your team has the tools), permission (explicitly tell them you expect and encourage AI use), and a feedback loop (create one regular moment where people share what they learned). A single 15-minute team ritual around AI sharing, sustained for 90 days, will do more for your team's AI capability than a two-day offsite workshop that happens once.

Click to enlarge

Key Lessons From This Section

1. AI skills develop through real work, not just formal training, build learning into daily workflows rather than relying on scheduled programs.
2. Peer sharing accelerates AI fluency faster than almost any other mechanism, create structured moments for teams to exchange what's working.
3. Different roles need different AI skills, a recruiter needs bias-detection judgment, a salesperson needs tone calibration, an executive needs delegation sense.
4. Psychological safety is a prerequisite, teams that can share what didn't work with AI learn faster than teams where failure feels risky.
5. Leadership modeling matters enormously, when managers openly use and discuss AI, permission spreads through the team.
6. Organizational AI capability compounds over time, the teams that start building shared fluency now will have significant structural advantages within 12–18 months.
7. Access plus permission plus a feedback loop is the minimum viable AI culture, you can start all three this week with no budget.

Takeaways

• The best AI talent strategy is a culture of low-stakes experimentation woven into real work, not a standalone training curriculum.
• Walmart, Etsy, and Unilever each demonstrated that context-specific, peer-reinforced learning outperforms generic AI workshops.
• A simple weekly sharing ritual, 3 to 5 minutes, one person, one AI use case, compounds into significant team capability over 90 days.
• Skills gaps are role-specific: identify what each function actually needs from AI, then develop toward that target.
• The transition from individual AI use to organizational AI capability is a leadership responsibility, and it starts with modeling, access, and one feedback loop.