The problem

Human coaching offers depth but not scale, so it mostly reaches a select group of leaders. Learning platforms reach everyone but deliver content without clarity or continuity. Most employees fall in the gap: they want to grow but don't know where to start, and they rarely have time or structure to reflect on feedback.

We saw the gap inside our own product. Many learners struggled to complete a coaching plan without a coach, so coaches spent valuable session time on paperwork instead of coaching. Customers were asking for AI coaching, and competitors were moving fast. A generic chatbot would not compete; dedicated coaching platforms win on trust, continuity, and depth.

The design challenge: create an AI that behaves like a good coach (curious, non-directive, and safe) inside a platform built for learning content, without blurring the two.

Strategy: one coaching architecture

Skillsoft was building several AI agents across the platform at once. Left alone, learners would face a collection of disconnected bots. I authored the AI agent strategy that made them one experience: an orchestrator routes each request to the right specialist, agents share context, and every exchange ends with a next step.

Three principles shaped the strategy:

• Learning and coaching are different modes. Learning is about discovery and consumption; coaching is about reflection, sense-making, and accountability. I kept AI Coach a focused space instead of folding it into the learning feed.

• Clear ownership of data. Learning owns skills data; coaching owns goals, reflection, and behavior change. No parallel skill scores, so definitions never drift.

• Skills become behaviors. A skill gap like "strategic communication" becomes real-world experiments: lead your next meeting with outcome framing, then reflect on how it went.

The architecture works for every learner. Learners without skills data get the same core coaching; learners with it get added personalization, not a different product.

Designing how the AI shows up

The hardest design problem wasn't a screen. It was behavior. A prompt shapes one answer; I designed the meta-behavior, the system-level rules that govern how the AI Coach shows up over time. Meta-behavior was UX-owned, and I wrote it to align with International Coaching Federation (ICF) competencies so the AI behaves like a coach, not an advice engine.

• Coaching agreements. The AI starts with collaborative goal-setting, rechecks the session's focus each time, and gently redirects when a conversation drifts. It tracks long-term goals and session goals separately.

• Trust and safety. Nonjudgmental reflections; validating feelings without over-identifying. Tone adapts to sentiment: when a learner expresses distress, the AI shifts to a grounding, gentler tone.

• Presence and pacing. It mirrors the learner's language, paraphrases precisely, and keeps a calm, spacious pace that supports reflection rather than urgency.

• Clear boundaries. It never positions itself as an expert, therapist, or authority. Medical, therapy, and HR-policy topics are redirected to the right human resources.

Voice: supportive, calm, encouraging, clear, and professional. Never judgmental, prescriptive, overly casual, or like a help-desk bot.

I designed meta-behavior to be explainable, auditable, and configurable, so the AI Coach can grow with the learner while staying consistent.

The experience

Each screen puts one design decision into practice.

1. A welcome that sets expectations

The AI Coach opens by explaining what it is and what it isn't: personalized goal setting, progress tracking, insights, and reflection, grounded in ICF coaching principles. Setting expectations up front builds trust before the first question.

2. Getting to know you, then a self-assessment

Onboarding asks about role, responsibilities, growth priorities, and motivation, then runs a short self-assessment. Every learner gets the same core onboarding; learners with Percipio skills data simply get added personalization.

3. The first goal conversation

The AI offers focus areas as chips to lower the effort of starting, then asks open, non-leading questions: "What is one situation where you would like to show up differently?" It reflects back what it hears and offers to turn the answer into a goal, with options to refine it or identify tasks. A side panel shows where the learner is in the journey.

4. Goals become actions

Goals break into tasks and suggested resources, so reflection turns into real-world practice. Open, completed, and empty states were all designed so the page stays motivating at every stage.

5. Bringing the manager in

Learners can request feedback on their goals, and managers respond through a lightweight pulse check. This closes the loop between coaching and real work, and gives learners outside perspective without a heavy review process.

6. A home base to return to

The home screen invites the learner back with their current goal, next task, and recent conversation, plus coaching moments like preparing for a difficult conversation. "We protect what you share" sits beside every input, because trust is the product.

Outcomes

• 2024: AI Coaching Guide MVP shipped (July). It's available 24/7 and helps learners build a SMART-goal coaching plan between sessions, so coaches can spend session time on real coaching.

• 2026: AI Coach. Launched as a premium offering in Skillsoft's Percipio platform, built on the orchestration strategy and behavior framework I authored. It has driven substantial new revenue.

What I learned

• Behavior is the interface. In AI products, the most important design work happens in rules, tone, and boundaries, long before pixels.

• Restraint builds trust. Deciding what the AI should not do (diagnose, prescribe, score) mattered as much as what it does.

• Separate modes, shared context. Keeping coaching distinct from learning, while sharing data underneath, protects both experiences.