Bojan Savic recently shared something that caught my attention: "Valentine's Day. The perfect time to talk about mending the SME-ID relationship." Then he dropped a painfully familiar exchange:

SME: "The API handshake times out."
ID: "...can you explain that like I'm five?"

Bojan summed it up in one line that will make a lot of learning teams nod: "Knowledge gets lost in translation. Every. Single. Time." I have lived that loop, and I bet you have too. The subject matter expert (SME) speaks in compressed expert shorthand, the instructional designer (ID) tries to decode it, and what should have been a "quick sync" turns into five meetings, three half-finished docs, and a course that still misses key details.

In his post, Bojan points to an emerging fix: an AI layer that captures SME knowledge and turns it into learner-ready content without the endless interview loops and rewrites. Let me expand on why this gap exists, what it costs, and what "closing the translation gap" can look like in practice (with and without AI).

The real problem is not expertise, it is translation

Bojan describes a loop that repeats across industries:

SME speaks in expert shorthand
ID scrambles to translate
"Let me clarify" becomes follow-up meetings
Knowledge gets lost, time gets wasted, everyone is frustrated

What is actually happening is a mismatch in "compression." SMEs compress because they can. Years of pattern recognition let them pack meaning into a few words. IDs and L&D teams, on the other hand, must expand meaning for a learner audience. Expansion requires context: definitions, assumptions, examples, edge cases, and decision criteria.

So when an SME says "The API handshake times out," they might mean:

Which system talks to which system?
What authentication method is involved?
What is the expected latency and the failure threshold?
What does a timeout look like in logs?
What action should the learner take when it happens?

The SME has those answers in their head, but the ID cannot design effective practice activities until those answers are explicit.

Why "communication guides" rarely fix it

Bojan jokes (accurately) that we make guides, and the gap stays wide. I have seen plenty of "SME interview question lists" and "how to work with SMEs" playbooks. They help, but they do not eliminate the core friction because the bottleneck is not knowing what to ask. The bottleneck is the repeated, manual conversion of tacit knowledge into structured learning artifacts.

Even with a good guide, the process often looks like this:

The ID collects fragments (meeting notes, screenshots, Slack explanations).
The ID drafts a storyboard based on interpretation.
The SME reviews and says: "Not quite, it is more like..."
The ID revises, introduces new errors, or loses nuance.
Another review cycle begins.

This is not anyone failing. It is a system problem: knowledge transfer is treated as a series of interviews rather than a productized pipeline.

The hidden cost of the SME-ID loop

Bojan mentions "3 hours in a 'quick sync' that went nowhere." Multiply that across a program and you get:

Longer cycle times (weeks become months)
Lower SME engagement (they start avoiding reviews)
Lower design confidence (IDs ship safer, less specific learning)
More rework after launch (support tickets become feedback)

The most expensive part is not the meeting. It is the downstream ambiguity: unclear performance outcomes, vague assessments, and content that teaches terminology instead of decision-making.

What it means to "close the gap" with an AI layer

Bojan highlights Powtoon's direction: "It captures SME knowledge and turns it into learner-ready content." That phrase matters because it implies something beyond transcription.

A transcript is still SME language. Learner-ready content is structured and contextual. To be truly helpful, an AI layer has to do at least four jobs:

1) Elicit missing context automatically

Instead of waiting for the ID to notice gaps, the system should prompt for what is missing:

"What is the most common cause of this timeout?"
"What should a beginner check first?"
"What mistakes do new hires make here?"
"What is a realistic scenario where this occurs?"

This is where AI can shine, not by replacing the ID, but by acting like a relentless clarifier that never gets tired.

2) Translate into multiple levels of understanding

Bojan’s "explain that like I'm five" line is not about dumbing things down. It is about matching the explanation to the learner's mental model.

A useful workflow is producing three versions of the same idea:

Executive summary (what and why)
Practitioner guide (how and when)
Troubleshooting playbook (signals, causes, actions)

AI can draft these tiers quickly, then the ID selects the right one for the audience.

3) Convert knowledge into practice, not just content

Learners do not need more paragraphs. They need decisions, consequences, and feedback.

For example, from "API handshake times out," a learner-ready output might include:

A branching scenario: "The integration fails at 2am. What do you check first?"
A checklist job aid: logs to inspect, thresholds, escalation steps
A short assessment: identify the likely cause from symptoms

If AI helps generate plausible scenarios and distractors, it shortens the distance from SME knowledge to usable practice.

4) Preserve traceability back to the SME source

One reason knowledge gets lost is that edits erase provenance. A strong system keeps a clear link between:

SME claim (source statement)
ID interpretation (learning objective)
Final asset (script, slide, microlearning, quiz)

When the SME disagrees, you can pinpoint the translation step that introduced drift.

A practical "relationship repair kit" that does not require new tools

Bojan’s giveaway idea (including actual LEGOs) is funny because it is true: rebuilding the relationship is often about rebuilding the workflow. Even if you do not have an AI layer yet, you can reduce friction with a few repeatable moves.

Use "decision-first" prompts in every SME session

Instead of "Tell me about X," ask:

"What decision should the learner be able to make?"
"What information do they need to make it correctly?"
"What is the most common wrong decision, and why?"

This naturally pulls out criteria, which is what IDs need.

Build a shared glossary that includes examples

A glossary is not definitions. It is:

Term
What it means in this context
One correct example
One near-miss example

That last part is where the learning value lives.

Replace reviews with validations

Reviews invite broad feedback: "This feels off." Validations are specific:

"Is this scenario realistic? (yes/no)"
"Is option B a plausible mistake? (yes/no)"
"Would you do step 3 before step 4 in real life? (yes/no)"

This lowers SME time and increases accuracy.

Where AI helps, and where it can hurt

I agree with Bojan that the category of tool he describes matters. But it is worth naming the guardrails.

AI should reduce the number of translation cycles, not add a new layer of rework.

To make AI useful in SME-ID collaboration:

Require SME sign-off on critical claims (especially compliance, safety, security)
Keep prompts and outputs stored with version history
Use AI for structure, first drafts, and scenario generation, then have humans verify
Measure impact: fewer meetings, faster storyboard completion, fewer post-launch corrections

The risk is hallucinated specificity. If AI invents log messages, thresholds, or policies, it can create training that sounds confident and is wrong. The solution is not avoiding AI, but designing the workflow so that validation is built in.

The point Bojan is really making

Beyond the Valentine’s framing and the giveaway mechanics, Bojan is calling out a daily pain point in L&D: the SME-ID relationship breaks down when translation work is invisible and unproductized. When translation becomes a repeatable system (and ideally partially automated), both roles get to do higher-value work.

SMEs spend less time repeating themselves. IDs spend less time decoding jargon in real time. Learners get content that reflects reality, not a best guess.

If you have ever left a meeting with pages of notes and still thought, "I am not sure what they actually do," you already understand why a tool that closes this gap is worth paying attention to.

This blog post expands on a viral LinkedIn post by Bojan Savic. View the original LinkedIn post →