​AI agents exploded by adding more scaffolding around models: Planning systems. Retrieval pipelines. Memory layers. Reflection loops. Tool orchestration. Multi-agent workflows.

​But frontier models are starting to absorb many of those capabilities directly.

​The result?
A lot of today’s “best practices” for building agents may already be temporary.

​In this live conversation, Hugo Bowne-Anderson and Nicolay Gerold (Amp Code) will explore what’s actually changing in agent engineering right now: what still matters, what’s collapsing into the model itself, and how builders should adapt.

​This is a practical, technical conversation for people building agents, coding tools, AI products, and LLM systems today.

​What We’ll Cover

• ​Why newer models often break older agent harnesses

• ​The shift from prompt engineering to context engineering

• ​Why planning, memory, and orchestration are increasingly moving into the model

• ​What coding-agent infrastructure actually looks like in production

• ​Why retrieval and context management matter more than ever

• ​The tradeoffs between recursive systems and simpler agent architectures

• ​Why many current agent abstractions may disappear

• ​What AI builders should still build as models get dramatically stronger

• ​How frontier teams evaluate new models and redesign systems around them

• ​Where durable engineering leverage still exists for agent builders

​Register to join live or get the recording afterwards.