A colleague recently complained to me about the hassle of organizing information in AGENTS.md / CLAUDE.md. This is the mark of a real adopter - she has gone through the progression from being impressed by coding agents to being annoyed at the next bottleneck.

When I'm thinking about optimizing repos for agents, I'm looking to accomplish three main goals[^1]:

• Increase iterative speed: Avoid repeated context gathering, enable the agent to quickly self-correct its mistakes.
• Improve adherence to evergreen instructions: Over time, repeated agent mistakes emerge. Context within the repo helps the agent avoid these and adopt a more consistent workflow.
• Help the most agentic agents of them all: Humans and agents scan docs and code in very similar ways, so organizing information so it's easily understood by humans is a good rule of thumb for helping the agents anyways!

A reshaped learning curve​

Before AI, learners faced a matching problem: learning resources have to be created with a target audience in mind. This means as a consumer, learning resources were suboptimal fits for you:

• You're a newbie at $topic_of_interest, but have knowledge in related topic $related_topic. But finding learning resources that teach $topic_of_interest in terms of $related_topic is difficult.
• To effectively learn $topic_of_interest, you really need to learn prerequisite skill $prereq_skill. But as a beginner you don't know you should really learn $prereq_skill before learning $topic_of_interest.
• You have basic knowledge of $topic_of_interest, but have plateaued, and have difficulty finding the right resources for $intermediate_sticking_point

Roughly, acquiring mastery in a skill over time looks like this:

A core challenge of using LLM's to build reliable automation is calibrating how much autonomy to give to models.

Too much, and the program loses track of what it's supposed to be doing. Too little, and the program feels a bit too, well, ordinary[^1].

For LLM-based applications to be truly useful, they need predictability: While the free-text nature of LLMs means the range of acceptable outcomes is wider than with traditional programs, I still need consistent behavior: if I ask an AI personal assistant to create a calendar entry, I don't want it to order me a pizza instead.

While AI has changed a lot about how I develop software, one crusty old technique still helps me: tests.

This is a brief overview of my advice for new grads and junior software engineers. I'm been in the industry for about 8 years, and worked my way into engineering without a computer science degree. I've worked in both startups and medium-sized companies over the past 8 years.

As is the case with lots of tech writing, my advice will be skewed towards working in the San Francisco bay area, without needing visa sponsorship. Location and residency status are major factors to think about.

Other engineers with similar levels of experience as mine will disagree with some or all of it.

OpenAI launched its GPT Store this week. Brands and developers can create custom GPT's, either for sale or for free. Both have eagerly launched many GPT's, probably due to the relatively low overhead of creating them.

I am skeptical of the value. For brands, this feels like the AI equivalent of a service that sends postcards in response to an email. The access pattern and interface are more or less exactly the same as traditional apps or sites, only via a GPT. It's a neat trick, but I think users will quickly lose interest.

MemGPT is an interesting project which provides GPT agents with unbounded memory. It includes the ability to incorporate custom functions, with a convenient JSON schema generator.

In trying to extend the agent with functions of my own, I found that the agent was reluctant to give me information about the functions I was making available to it, so I wrote a set of meta-functions which enable the agent to view source code, set debugger lines, and create functions. You can view the source code here. Note that running this requires some edits I made to MemGPT to enable dynamic function reloading (PR).