The Gap Between Individual Fluency and Organisational Capability¶
A few weeks ago I asked a room of developers two questions.
First: "How do you use AI in your work?"
Great answers. Specific. Confident. Everyone had a workflow.
Second: "What does your team's AI workflow look like — the shared one?"
Silence.
That silence is diagnostic. It's not that the team wasn't using AI — they were. It's that the capability lived in individuals, not in the organisation. One developer had figured out a systematic approach. The rest watched over his shoulder. That's not a team workflow. That's a dependency.
Two terms, one gap¶
The conversation right now is dominated by two terms.
Vibe coding describes the intuitive, feel-your-way-through-it mode that AI enables. You work in flow with the model — prompting, iterating, shaping. It's genuinely fast. It works. But it's tacit. It lives in one person's intuition. Ask that developer to explain their workflow and they struggle. Ask the person next to them to replicate it and they can't. It walks out the door when the developer does.
Agentic engineering is the other end — delegating sequences of tasks to autonomous AI systems with minimal intervention. Also powerful, especially for well-defined workflows. But it still doesn't answer the organisational question. The capability sits in whoever designed the agent setup, not in the team.
Both are real. Both produce results. Neither has a good answer to what was missing in that silent room.
What the best teams are building¶
What I'm watching emerge in the teams that are pulling ahead is something more deliberate than vibe coding and more human than agentic engineering.
Shared vocabulary. Documented skills. Repeatable workflows. Patterns that new people can learn and experienced people can improve. Human expertise setting the direction — AI in execution.
Capability that doesn't depend on which three people happen to be in the office today.
We've been calling it Skill-Driven Development. The name came from Vidar Moe at SpareBank 1, who saw it in practice and gave it a name before we had one.
What the comment thread surfaced¶
The most interesting response came from Freja AI — a Danish AI practitioner who works on sales team workflows. She pushed back on a specific assumption.
Her question: structured files work for individuals, but do they scale to teams? The problem she'd seen was the gap between documented and internalized — and the unclear ownership that comes when a shared workflow is nobody's job to maintain.
She named a harder version of the problem: unknown unknowns. Teams that don't know what they know don't know what to search for. If the only path to your organization's AI capability is through a search interface, you've only solved the easy case — the knowledge people already know to ask about.
That's a fair critique and a sharp one.
The distinction I drew back: Synthesis and SDD are solving different things. Synthesis addresses the known → findable problem: taking structured knowledge that exists and making it retrievable. SDD addresses the tacit → explicit problem: taking knowledge that's currently locked in individuals' heads and externalizing it into documented, transferable skills.
The unknown unknowns she described — the knowledge the team doesn't know it has — are a third category. And her answer for those was interesting: they shrink through practice, not through better search. The self-learning loop. Teams that run deliberate skill composition sessions surface unknown unknowns naturally, because the process of making tacit knowledge explicit forces the articulation of things that were never written down.
She'd already been doing this herself, using SKILL.md files to document personal workflows. She just hadn't called it SDD.
The diagnostic question¶
Michael Åhs, who runs AgentBrew Oslo, sent a DM within three hours of the post going up. He'd recognized the silence moment from enterprise clients. Same gap, different entry point.
That's the thing about the silence in the room: it's not rare. Most teams, when pressed on their shared AI workflow, produce exactly that. Not because they're behind — but because no one has asked the question before.
The question is what you do with the answer.
If the honest response is "our best developers have their own approaches and the rest follow along," that's a dependency structure, not a capability structure. It produces results today. It's fragile to turnover. It doesn't compound.
If the goal is capability that compounds — that gets better as the team grows, survives the people who built it, and trains the next hire systematically — the path is SDD: deliberate, shared, documented, human-led.
The gap isn't between organisations using AI and organisations that aren't.
It's between organisations that know which direction they're drifting — and those that don't.
Originally published on LinkedIn, February 20, 2026.
This post is part of the AI-Augmented Development series.