unknown

I think this is a likely evolution. For sure building agentic software using Claude Agents SDK or DeepAgents SDK really works well. 

And there's clearly a compounding effect if you are able to manifest learnings of your agent sessions into context that is useful to the agent on the next run. 

There are many ways to slice the onion here, but I think self-updatable software in many verticals which automatically captures some essence of user preferences is going to happen for sure. 

In fact all the building blocks are there such that it can exist already - I think we just don't have an 'aligned set of recipes / patterns' that are established and used consistently by software teams. But we'll probably get there soon, and honestly there is a TON of opportunity here to figure out the right ways to do it. 

Systems like Claude Code / Deep Agents / Claude Agents SDK are just a set of legos. So many ways to put it together.

<p>I think this is a likely evolution. For sure building agentic software using Claude Agents SDK or DeepAgents SDK really works well. 

And there's clearly a compounding effect if you are able to manifest learnings of your agent sessions into context that is useful to the agent on the next run. 

There are many ways to slice the onion here, but I think self-updatable software in many verticals which automatically captures some essence of user preferences is going to happen for sure. 

In fact all the building blocks are there such that it can exist already - I think we just don't have an 'aligned set of recipes / patterns' that are established and used consistently by software teams. But we'll probably get there soon, and honestly there is a TON of opportunity here to figure out the right ways to do it. 

Systems like Claude Code / Deep Agents / Claude Agents SDK are just a set of legos. So many ways to put it together.</p>

Harrison Chase

A retweet expressing agreement that building agentic software using Claude Agents SDK or DeepAgents SDK is a likely evolution in software development.