← Back

Talent 2030

Jon Grinnell · June 8, 2026 · 9 min read

Intro

I work in learning and development (L&D), and I spend quite a bit of time using state-of-the-art (SOTA) AI tools in my role. Lately, I've been asking myself:

“What could learning and performance at work look like in the year 2030?”

If one looks at the recent developments of agentic AI, and what’s going on thematically with tools like OpenClaw, GStack, Claude Cowork, Gemini Spark, etc., then we can gain a directional sense of where things are headed in terms of AI at work. And of course, this will impact talent management within companies. In my observation, two themes are emerging that suggest how AI might materially change the way we learn and perform at work:

1. Structured context > better LLMs. If AI has enough structured information on your current work, across all your tools and systems, it helps the AI make better decisions for you. In many cases, this is more important than the LLM model itself, as models are converging in terms of quality and capabilities. Think about how many tools and systems the modern knowledge worker uses at work: Slack, Gmail, Docs, Sheets, Zoom, Meet, Asana, Salesforce, MS Office Suite tools, Teams, etc. It’s not common today to connect all the systems together and have a holistic contextual view of how your work is connected across your apps and tasks. With a higher awareness of your work, your approach, and your patterns, context is becoming the differentiator.
2. AI is already capable of learning. AI systems are already able to autonomously log and store new memories in way that feels like the AI *gets you*. Take any SOTA AI model: Gemini, Claude, ChatGPT, Grok. They all prioritize memory functionality because it is what makes the product *feel* more personalized to you. Gemini has personal intelligence so it learns from your Gmail, Google Calendar, GDrive, etc.. Claude Code has /autodream to optimize and prune memory automatically. They all allow you to import memories from another provider. With structured context and better memory systems, these agents and subagents, often working in parallel, learn and remember in increasingly useful ways.

The concept

**Talent2030** can be thought of as a research project on what the future of learning and performance could look like from the employee lens. I felt it would be interesting to imagine a fictitious AI company called **Acme AI Co.** (bear with me on the unremarkable placeholder company name). In this company, users are logged in to the company talent portal as Will Ray, an imaginary VP of Product at Acme AI Co. who manages a team of engineers, designers, and analysts. Will is approaching performance reviews for his team and has a tricky situation with Mark, who is underperforming.

In most AI platforms today, if Will wanted to consult AI and strategize how to approach the conversation, he might have to give the AI a ton of information. For example, Will’s prompt might read:

I'm approaching a conversation with Mark, a product manager on my team, and I need to deliver his performance review. I'm feeling an underprepared and the performance review process I need to follow is too dense so make it simple. See attachments:

- our performance review process on the Wiki page 
- some Slack conversations I've had with Mark
- reviews from Workday
- some of the observations I've had on his work
- some meeting transcripts with Mark

how should I approach this?

This prompt could be quite helpful. It could help Will gain an initial perspective on how to approach the conversation with Mark. But let’s assume Will has other, time-sensitive work priorities as the VP of Product. Maybe he also has a situation at home that he keeps thinking about. Maybe the Slack examples he chose from Mark are cherry picked, maybe the meeting transcripts unintentionally suggest that Will himself has some leadership skill gaps. Or maybe, just maybe, the Confluence link to the performance review process is the wrong one, outdated from 11 months ago with an inactive HR employee as the author. Yikes. Now, the AI is advising the wrong process, with low-signal context, just as Will’s been pulled into a Code Red on a product issue. Now, let’s ask the question:

**What if the agent already knew all of this?**

Well, that would be convenient, right? Instead of a *cold start*, if the agent already knew what was going on with Mark, it could come to Will preloaded with all this insight to help him navigate and even simulate this conversation with Mark. Furthermore, if the agent had a canonical, corporate-approved and clearly articulated performance management document, it could use that in its coaching approach. This means that Will can know exactly where in the perf cycle he is, and feel more enabled going into reviews. Talent leaders provide the process document, and conversational agents can provide an interaction layer for Will to enact the process. And that’s what I built.

The product

[Try the live demo →](https://talent2030.vercel.app/) — explore Acme AI Co. as Will Ray, VP of Product.

Will Ray, the VP of Product, has an AI coach that knows who he is in the company, his goals, his team, his workflows, the tools he uses, and more. To do this, the agent uses an `employee.md` file, a structured context document that automatically updates based on Will’s tasks, his team, and anything he does at work. Think of this as a structured ledger on Will’s work activity that grows, prunes itself, and evolves. In addition, the agentic coach also uses `talentmanagement.md`, a document provided by talent execs from Acme AI Co.. Think of this as a one-pager that outlines the performance process in plain language.

employee.md

# Employee context — Will Ray

Agent context file for the **manager using Talent Management** (not a direct report file). This profile updates in real time as the manager uses the app.

## Profile

- **Name:** Will Ray
- **Title:** VP of Product
- **Company:** Acme AI Co.
- **Is people manager:** Yes — 5 direct reports
- **Review cycle:** 2026 Annual Performance Review (Delivery meetings)
- **Delivery window:** May 19 – Jun 6, 2026

## Team

### Mark Webb

- **Role:** Senior Product Manager (L5)
- **Tenure:** 2 years 4 months
- **Status:** PIP watch
- **Signals:** Checkout refresh slipped in Q1; completion never improved. Defensive in Slack when challenged on scope. 1:1 follow-through slipping.

### Priya Nair

- **Role:** Staff Engineer (L6)
- **Tenure:** 4 years 1 month
- **Status:** Promotion track
- **Signals:** Led platform migration ahead of schedule. Mentors two engineers. Strong calibration case for principal track.

### Jordan Lee

- **Role:** Product Designer (L4)
- **Tenure:** 1 year 8 months
- **Status:** On track
- **Signals:** Solid craft and collaboration. Still building end-to-end ownership on the Growth product group.

### Alex Rivera

- **Role:** Data Analyst (L3)
- **Tenure:** 6 months
- **Status:** 6-month check-in
- **Signals:** Strong analytical foundation at six months. Still ramping on stakeholder communication and prioritization.

### Taylor Brooks

- **Role:** Program Manager (L5)
- **Tenure:** 5 years 3 months
- **Status:** Retention risk
- **Signals:** Reliable operator across three launches. Has flagged compensation and scope concerns in recent 1:1s.

## Tools used

- Slack
- Gmail
- Google Meet
- Google Drive
- Calendar
- Asana
- Workday

## Learning and interests

- AI product strategy and responsible deployment
- Executive communication for difficult conversations

## Collaboration patterns

- Weekly 1:1s with all five direct reports
- Product leadership channel (#product-leads) — daily async
- Calibration prep with VP Engineering and People Ops
- Cross-functional partner: Priya Nair (platform), Jordan Lee (design)

## Activity in Talent Management

Recent actions in this demo (newest first). Use for continuity—reference lightly, do not read verbatim.

- Jun 4, 2026, 12:34 PM — [Team] Reviewed context for Mark Webb — Senior Product Manager (L5) — PIP watch. Browsed self-review, manager review, and workplace signals.
- Jun 4, 2026, 12:32 PM — [Team] Reviewed context for Mark Webb — Senior Product Manager (L5) — PIP watch. Browsed self-review, manager review, and workplace signals.

talentmanagement.md

# Talent management: Acme AI Co.

Agent context file describing how performance management works at Acme AI Co.. Use this as organizational ground truth for coaching, planning, and review delivery.

## Organization

**Mission:** Build inference infrastructure that helps enterprises deploy AI safely at scale.

**Talent philosophy:** Learning-forward culture with evidence-based feedback. Managers act as coaches, not judges. Performance conversations should develop people, clarify expectations, and reinforce values: rigor, ownership, and customer impact.

**Review cadence:** Annual formal performance reviews with mid-year check-ins. The 2026 annual cycle covers self-review, calibration, delivery meetings, and Workday documentation.

**Official policy:** [Performance Management Process](https://acmeai.atlassian.net/wiki/spaces/People/pages/884742/Performance+Management+Process) (People & Talent, updated Mar 12, 2026)

## Current review cycle

**2026 Annual Performance Review**. Current phase: Delivery meetings

- **Cycle kickoff** (Mar 24 – Mar 28): People Ops publishes guidance. Managers align teams on timeline and expectations. (complete)
- **Self-review** (Mar 31 – Apr 15): Employees reflect on goals, impact, and development areas in Workday. (complete)
- **Manager draft** (Apr 16 – May 10): Managers write rating rationale, evidence, and next-year goals. (complete)
- **Calibration** (May 12 – May 16): Leadership normalizes ratings across teams. Document any changes and rationale. (complete)
- **Delivery meetings** (May 19 – Jun 6): Dedicated 45–60 min conversations to deliver rating, feedback, and co-create goals. (in progress)
- **Workday close-out** (Jun 9 – Jun 13): Final rating, summary, and goals entered within 48 hours of each meeting. Employee acknowledges.

## Performance management process

1. **Self-review** (employee, ~2 weeks): Employee reflects on goals, impact, and development areas in Workday. Manager should read before drafting.

2. **Manager draft** (~1 week): Manager writes rating rationale, evidence, and next-year goals. Must align with calibration guidance.

3. **Calibration** (leadership): Managers present cases; ratings are normalized across teams. Document any rating changes and rationale.

4. **Delivery meeting** (45–60 min): Manager delivers rating and feedback in a dedicated conversation, not as a surprise in a status meeting.

5. **Workday documentation** (within 48 hours): Final rating, summary, and goals entered. Employee acknowledges in system.

**Manager responsibilities:** prepare with evidence, create space for employee voice, co-create goals, and document accurately.

## Rating scale (annual review)

Acme AI Co. uses a **1–5 scale** in Workday. Ratings must be supported by evidence and calibration.

| Rating | Label | Guidance |
|--------|--------|----------|
| 1/5 | Significantly below expectations | Sustained misses on core commitments; immediate improvement plan required; partner with HR. |
| 2/5 | Below expectations | Missed goals or material gaps; clear recovery plan and frequent check-ins. |
| 3/5 | Meets expectations | Delivers role expectations reliably; balanced strengths and growth areas. |
| 4/5 | Exceeds expectations | Consistent high impact beyond role; strong evidence across multiple quarters. |
| 5/5 | Significantly exceeds expectations | Exceptional, org-visible impact; often promotion or scope expansion case. |

**Calibration note:** Ratings are normalized across teams. Managers must explain the "why" with 2–3 anchor examples at delivery.

## Successful review meeting

**Opening (5 min)**
- Set intent: this is their review, not a project update.
- Share agenda: their self-reflection, your feedback, rating, goals, support.

**Evidence-based feedback (20–25 min)**
- Use **COIN** (Context–Observation–Impact–Next steps): specific examples, observable behavior, business impact, and agreed next steps.
- Balance strengths and gaps. Tie feedback to role expectations and level.
- Invite their perspective before moving on. Ask: "What resonates? What feels off?"

**Rating delivery (5–10 min)**
- State rating clearly. Explain the "why" with 2–3 anchor examples.
- No new critical feedback at rating reveal. Nothing they haven't heard before.

**Forward look (10–15 min)**
- Co-create 2–3 measurable goals for the next year.
- Discuss growth, support, and resources.

**Close (5 min)**
- Summarize commitments on both sides. Confirm next check-in.

**Anti-patterns to flag**
- Vague praise ("great attitude") without examples
- Monologue without pauses for employee input
- Surprise negative rating or new serious concerns
- Comparisons to peers by name
- Promises you cannot keep (guaranteed promotion, compensation)

## Underperformance handling

**Early signals:** missed commitments, quality gaps, disengagement, or conflict avoidance. Address in 1:1s with clarity; document themes.

**Structured improvement:** clear written expectations, weekly or biweekly check-ins, offer support (coaching, reduced scope, pairing).

**PIP thresholds:** two consecutive quarters below expectations, or one quarter significantly below with no credible recovery plan. PIP typically 30–60 days with weekly documented check-ins. Partner with HR before initiating.

**Conversation tone:** direct and specific, not punitive. Separate intent from impact.

## Connected systems

Assemble employee and team context from:

- **Slack**: tone, collaboration, team dynamics
- **Gmail**: written communication and stakeholder threads
- **Google Meet**: meeting behavior and delivery patterns
- **Google Drive**: work artifacts and document quality
- **Calendar**: 1:1 cadence and follow-through
- **Asana**: commitments, ownership, and delivery tracking
- **Workday**: self-reviews, ratings, goals, and org structure

The canvas

Will interacts with the agentic coach through an **audio-visual canvas**: a live agent that generates conversational audio paired with a visual user interface in real-time. He can use the manager copilot canvas, offering him leadership coaching, or a learning canvas, which offers personalized microlearning. Here’s what the manager copilot canvas looks like:

Video: Manager Copilot — /videos/manager-copilot.mov

I’ve been in L&D forever, so I also wanted to reimagine the learning process with this project. That’s the Guided Skill path section in the demo. When selecting a skill path, instead of opening a static course, it kicks off with a compressed consultation (3 quick questions) to understand what the already knows, what he wants to learn, and his preference on how he wants to learn it. For example, Claude Code. If Will wants to learn about Claude Code to be more informed with his engineers (like Priya - see the "Team" tab), it will guide him through that. The agent also reads the `employee.md` file to profile Will before building the path, meaning it accounts for his recent activity, his role, tools he already uses, any relevant context that makes the learning hyper-relevant. And by design, it only outputs 5 slides, all text, all grounded on current information on the web and the business. Will can pause at any point, ask questions out loud, and go down rabbit holes if he would so choose. Most critically, it supports non-linear learning while still retaining a linear structure. In the end there’s a knowledge check, which when complete, logs as a lesson learned in `employee.md`. For the learning nerds, the canvas simulates adaptive learning, the `employee.md` file serves as an LRS, and the skill path creation simulates a SAM instructional design methodology. However, there is no LMS, no SCORM courses, no ID tools, just contextual, fast microlearning.

Video: Guided Skill Paths — /videos/claude-code.mov

Team & rehearsal simulation

In the Team tab, Mark has visibility into his team, should he want to see their written self-reviews, his written reviews, or agentic recommendations for delivering the performance review based on several context-rich sources.

The product also has the ability to simulate a performance review conversation with Mark via a Google Meet style virtual call. Will can also see a heads-up display (HUD) for tips or talking points along the way, optionally. This is a chance for Will to actually practice the conversation before it happens. In our example, the simulation also has full context on Mark's performance (what he rated himself, what Will rated him, where he is in the process, and frameworks to apply).

You might find that in the simulation, Mark can be a bit ... spicy. In my view that can be a good thing. It's a chance for Will to practice a difficult conversation that requires range, how to stay cool when things feel heated, and also how to ensure Mark feels heard and valued even if he is defensive. I encourage you to read Mark's self-review and Will's manager review to get a sense of his performance, and see how you perform in this scenario. In the end, Will has the option to assess himself on how he performed and delivered (6:03 in the video).

Video: Team & rehearsal simulation — /videos/convo-final.mov

Product design choices

I made intentional design choices in the product, two in particular.

1. Voice is the future. We speak 3.1x faster than we type and audio dictation tools have changed how I and many others use computers and phones. Fast, indistinguishable-from-human conversational AI with emotional expression is already here. For example, Will has the option to effectively simulate that difficult conversation with Mark in a way that feels mostly authentic, in my view. By design, it’s cameras off, strictly audio. I chose not to use any live video, because I find that live AI avatars tend to be more distracting than helpful (for some reason we as humans are incredibly good at detecting any kind of fake video, and something about it is off-putting to our sensory perceptions. RIP Sora.)
2. The fixed canvas. Computers will increasingly have the ability to generate UI on the fly, likely directed by our voice. For instance, as Will uses the manager co-pilot feature and asks “how does our perf process work, again?”, the agent will both verbally speak and visually display a response. The canvas is designed for speed, without any whizbang animations or impressive fonts. Just text, to the point, reinforcing what the agent is verbalizing. Low latency audio paired with low latency, on-demand visuals.

Open questions & risks

There are many open questions as we think about context-aware, persistent AI at work. One of course is privacy. Do we really want an AI to see everything we do? In good faith, my sense is that most (not all) companies will do the right thing and enact guardrails that empower talent and provide transparency, vs. subtly spy on them, for lack of a better phrase. The other open question is around AI related job loss. Will AI just replace us? I think the pragmatic answer is that the degree of automation depends on the nature of the work itself. My view is that AI is mostly *middle-to-middle*, not quite *end-to-end*. There’s a human defining the work, AI does a bulk of the work in the middle, and a human validates the work in the end. But it does feel like AI will accomplish more work on the edges over time. Ultimately, it seems like there will be a distribution of tasks where AI will be mostly end-to-end without humans (for example, a call center support agent and progressively more complex knowledge work), whereas other higher risk tasks will usually require a human in the loop (for example - medical advice or surgery).

Closing & Outlook

In closing, this research project has sparked lots of new ideas for me in terms of skills and technology at work. It has illustrated how agents can scale talent outcomes if structured context is paired with conversationally fluid AI optimized for speed and impact. Lastly, as IQ is expected to scale exponentially, perhaps we will see companies invest more in [EQ](https://www.linkedin.com/posts/jongrinnellsf_joe-hudson-coaches-founders-at-openai-and-share-7447000097558790145-sGtw/?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAa_ZzYBZnvr6OjQWi3QN2X0mkgZyw6qmCg): how we navigate the emotional aspects of our work, how we relate to ourselves and others, and how we cultivate wisdom to make better decisions. After all, thats what really makes us *human*.