Same Engine, Different Transmission¶
March 4, 2026 · LinkedIn
9 reactions · 6 comments
Most AI productivity discussion asks the wrong question.
"How much faster?" assumes the difference is speed. It is not.
Two developers, same model, same subscription, same IDE. One starts every session cold. The other does not. The gap is not the engine. It is the transmission. And there is a third variable no one talks about: how the driver was trained.
The advantage I have built since January has two dimensions:
The infrastructure. Four memory layers where the standard setup has one. Working memory management (33% of context window recovered).
Episodic memory — every session indexed and searchable, including sub-agent sessions. Semantic memory across 9,000 files in three codebases. Two autonomous agents running overnight.
The methodology. Six practices called Skill-Driven Development that determine how fast the infrastructure compounds. Intelligent context, strategic delegation, trust but verify, directed synthesis, process discipline, continuous learning. Without the methodology, the infrastructure is tools without training.
The numbers by time horizon:
- Single session → 30–50% better
- Multi-session work → 1.5–3x
- Repeated domain work (5th engagement vs 1st) → 3–5x
- Mature domain, accumulated skills → not faster. A different category of achievable.
The last row is not a multiplier. It is a capability gap. Tasks that simply are not possible any other way within the time envelope.
Full post with slides and a downloadable PDF: [link in comments]
Where are you seeing the biggest gap — single-session efficiency, or multi-session continuity?
Discussion¶
Totto ↩: Post and slides: https://wiki.totto.org/blog/2026/03/04/same-engine-different-transmission/
Thor Henning Hetland: Thor Henning Hetland I'm using Claude Code + few custom skills.
Totto ↩: Sergiy Yevtushenko Perfect — kcp-commands slots right in alongside your custom skills, it doesn't replace them. It handles the tool layer (git, mvn, docker, npm, curl etc.) so your own skills can stay focused on your domain. Add it via kcp-mcp: npx -y kcp-mcp as an MCP server in Claude Code. Full setup in the README. 5 minutes. Let me know if you hit anything.
ObBonus: https://wiki.totto.or...