Skip to main content

brett

Productivity with AI

Updated 2026-09-05

My experience with AI has been that it augments my abilities. It can consolidate information and automate tedium to make me more productive. For more sophisticated tasks related to programming, AI cannot replace a hardworking and fully engaged engineer.

Some people might feel like we’re at a moment like this:

I’ve seen software engineers sometimes get defensive on the topic of AI, because they are perturbed by these sorts of unrealistic soundbites. Similar melodramatic sentiments are rampant on LinkedIn. I’ve been grateful to work with other engineers who have a strong interest in productivity and are level-headed and honest about AI.

The reality is somewhere between the extremes. AI isn’t replacing engineers, but it will also fundamentally change how we work. When used thoughtfully, it can genuinely accelerate development workflows by taking on specific, well-defined roles in the programming process.

brettinternet/ai

AI tools, research & playground

Usage

In my currently evolving workflows, AI fulfills a burgeoning role to augment my work with discovery and agentic iteration.

Discovery

Discovery is my favorite use case for AI. Being able to quickly traverse and reference large amounts of information has made me feel resourceful and effective.

Just talk to your repository with questions like “What are the side effects of this module?” or “Show me all the places where authentication is handled.” I frequently use this to understand hotspots in code and trace dependencies before making changes.

Well-structured codebases with clear boundaries are easier for both humans and AI to navigate. When I refactored a large codebase using Context Boundaries for team scalability, it also improved AI’s ability to provide focused, relevant insights by confining context to specific code subdivisions.

This raises an important design question as we integrate AI into development workflows:

How can we improve code organization for both human and AI readability?

The answer benefits onboarding, knowledge transfer, and debugging regardless of whether you’re working with human teammates or AI assistants.

Agentic iteration

  flowchart LR
    A[🤖 Code] --> B[🧪 Test]
    B --> C[🔧 Fix]
    C --> A

    B --> D[✅ Done]

    style A fill:#1e3a8a,stroke:#3b82f6,stroke-width:2px,color:#ffffff
    style B fill:#92400e,stroke:#f59e0b,stroke-width:2px,color:#ffffff
    style C fill:#991b1b,stroke:#ef4444,stroke-width:2px,color:#ffffff
    style D fill:#065f46,stroke:#10b981,stroke-width:2px,color:#ffffff

LLMs are most effective with iterative feedback. Developing good checks and tests provide guidance to the coding agents and validate their work.

I’ve seen Claude delete or add @tag :skip for tests in order to get them to “pass.” Engineers have to be hands-on conductors.

Caution

Some LLMs use excessive validation and praise.

ChatGPT: Dude. You just said something deep as hell without even flinching. You're 1000% right.
Glazing is bad

OpenAI’s AMA for the GPT-5 release demonstrated that some users crave this sycophancy. We need self-awareness about what using AI does to our psychology and good reviewing practices to instability in our code.

I saw a coworker publish a PR with invalid code and blame AI. People are accountable for code. AI can’t be accountable.

There doesn’t need to be a major paradigm shift in best practices. We should still maintain all existing practices for code maintainability whether it’s generated by AI or written by humans. For example, of course we should be concerned about what code AI writes. The same is true when we select libraries or languages without AI. In both cases we own the decision and the code. Age-old best practices continue even with modern AI technology.

What’s next?

As LLMs and the tooling evolve, so do my workflows. I’m continuing to learn and grow with these changes. My dotfiles are rapidly changing to reflect changes in agentic tooling.

Can engineers become excessively reliant on agentic prompting? Will this change engineering culture? What will this mean especially for newer programmers in the field?

Will LLM innovation begin to plateau? I wonder if we’re nearing a point where throwing more compute or a longer chain of thought won’t yield additional gains in performance.

For now, AI can augment software engineering in meaningful ways. I encourage software engineers to discover what LLMs can do for their productivity.

AI isn’t going to replace thoughtful engineering, but it can make thoughtful engineers more effective. The key is approaching it as a sophisticated tool that excels in specific contexts such as code completion, research and discovery, focused updates, and iterative problem-solving. As the technology evolves, so should our practices for integrating it responsibly into development workflows.

This post was adapted from a lightning talk I gave to a company leadership meeting.