Not tricks. The parts that things and feelings are made of.
Tokenomics
The talk takes apart the real economics of working with models: what a session costs (loading context, re-reading, generation) and which moves save money. On our metrics the spend falls to 13.2 billion tokens against a hypothetical 25.0 billion without the optimisations. The mechanics are shown layer by layer — from preparing the context to choosing a model class for the task; session profiles before/after and the return per token by month are included.
It suits ML/platform engineers, tech leads and product managers who scale AI in production and answer for the quality budget. It helps choose a model class for a task type, keep the patterns that work in the machine layer and judge the effect by numbers, not by feel.
Open →
Context management
The talk follows the path from local improvements of an AI editor to a multi-dimensional, measured system of context management. Quarter by quarter it sets out the milestones and the logic of choosing the vectors of the space; then a summary of the measurements against world practice and charts on data as of 2026-06-21 (AssetManager owns the facts). The final architecture of the loop closes the material, and the limits and the roadmap frame the next steps. The document stands on its own for reading and for saving as PDF.
It suits heads of product and platform, ML/data architects, CTOs and managers of an AI transformation. It helps align the language of decisions, compare metrics with world practice and justify development priorities without extra detail.
Open →
AI PDLC
The talk opens the delivery loop that is already running — from an idea to production through a kanban board, where an agent does the work and a person sets the gates. The path is 12 states across the board columns, tied to AIDLC/Agentic SDLC practice and to the implementation layers: architecture decisions, the experience layer, tests and end-to-end browser e2e. Historical milestones are shown by quarter; the facts are backed by report 007 (amendment 2026-06-20) and by the delivery-factory doctrine with no hand edits. The charts show how autonomous transitions and human gatekeepers are spread.
It suits product owners, tech leads and platform engineers who build or scale a delivery factory. It helps see the limits of automation, the points of human control and how the loop is proved by tracing back to sources. The limits of the current maturity and the roadmap of the next steps are stated.
Open →
