Since working with an LLM full-time for about a year now - and part-time for another 1-2 years, I've noticed my productivity vary along with specific work-practices; knowledge-worker practices; knowledge-management practices... from writing web applications, to drafting policy, to creating technical reports, and budget estimates, and slide decks, and visual charts, and animations, 3d assets, and video.

Certain practices are more congruent with AI than others.

Copy and paste is an anti-pattern.

Not persisting context is an anti-pattern.

Having LLMs generate more than an individual, team, or organization can handle (information overload).

Observations

• llm cli tools are powerful and flexible
• use placeholders with alt text that are llm prompts for the images
• focus on BDD (behavior driven development)
• work from/with git
• small steps (like good commit hygiene)

Pacing

• use caution
• work with intent

Thoughts

• llms allow one to recreate the wheel, faster than ever
• llms UI experiences are a huge opportunity - around collaboration, and personal file management and data management/ knowledge graph management -- there's something emerging here
• traceability and explainability will grow along with llm usage and adoption (a form of check and balance)