KOHSHEEN TIKU
ARTICLE · 32 MIN READ · FEBRUARY 19, 2026
There is no single AI alignment program — there's a portfolio. This chapter is the map: what each major research direction is trying to do, what its theory of impact is, what its current track record supports, and where it fits in the broader bet.
ARTICLE · 42 MIN READ · FEBRUARY 18, 2026
Real-world agents fail. Networks time out, APIs return errors, databases go down, data arrives malformed. This chapter is about building agents that handle every failure gracefully — detecting it, recovering from it, and keeping the user informed throughout.
ARTICLE · 36 MIN READ · FEBRUARY 14, 2026
Without goals, agents react. With goals, agents pursue. This chapter shows how to give AI agents specific objectives, measurable success criteria, and the feedback loops that keep them on track.
ARTICLE · 44 MIN READ · FEBRUARY 10, 2026
Every tool integration so far required custom code. MCP changes that — a universal standard that lets any LLM connect to any tool, database, or service without bespoke glue code for each pair.
ARTICLE · 39 MIN READ · FEBRUARY 06, 2026
Every pattern so far assumes the agent stays the same. Learning and adaptation break that assumption — agents that improve through experience, rewrite their own code, and discover better algorithms than humans designed.
ARTICLE · 35 MIN READ · FEBRUARY 02, 2026
Without memory, every conversation starts from zero. Memory management gives agents the ability to remember — short-term context within a session, long-term knowledge across sessions.
ARTICLE · 27 MIN READ · FEBRUARY 01, 2026
If a safety case is unfalsifiable, it isn't a safety case. This chapter takes the strongest critiques of the AI-safety program — accelerationist, infohazard-based, and interpretability-skeptical — and engages them on their own terms, before deciding which to update on.
ARTICLE · 39 MIN READ · JANUARY 29, 2026
One agent hits walls. A team of specialized agents doesn't. Multi-agent collaboration lets you decompose complex problems into coordinated workstreams — each agent doing what it does best.
ARTICLE · 37 MIN READ · JANUARY 25, 2026
Reacting to a single input is easy. Achieving a complex goal across many unknown steps is not. Planning is how agents develop foresight — decomposing a destination into a route before they move.
ARTICLE · 41 MIN READ · JANUARY 21, 2026
LLMs are frozen in time and disconnected from the world. Tool use breaks those walls — letting agents call APIs, run code, query databases, and trigger real actions.