AI as Coding Partner
AI-created, human-edited.
Harper Reed, former CTO of Obama's 2012 campaign and tech entrepreneur, joined Leo Laporte, Jeff Jarvis, and Paris Martineau on Intelligent Machines to discuss his current venture, which he described as building "super clippy or mega clippy" - AI agents that are "personified and a little bit unhinged" to help users with tasks.
Reed's appearance was prompted by his blog posts about using AI for coding, which caught Laporte's attention. During the conversation, Reed provided valuable insights into his workflow with AI coding tools, explaining how his team at his new company, 2389, now generates approximately 80% of their code using AI.
Reed outlined the progression of AI coding assistance as a continuum:
- Starting with traditional code autocomplete and IntelliSense
- Moving to advanced code completion tools like GitHub Copilot
- Advancing to pasting code sections into LLMs like Claude or ChatGPT
- Eventually reaching agentic tools like AIDR or Claude Code that can work autonomously
Reed emphasized that this progression requires users to develop both patience and prompting skills, noting that "if you start out saying it's not going to work, it's of course not going to work."
One of the most intriguing observations Reed shared was how AI coding is bringing back elements of the "waterfall" development methodology that Agile practices had largely replaced. He explained that effective AI code generation requires detailed specifications and planning upfront:
"To do well with LLM codegen and to really work in this space, you have to think about what the spec is for such a long time and so carefully and so strategically. You spend a lot of time working on requirements," Reed explained.
This represents a significant shift from recent programming paradigms, as developers must now focus extensively on requirements documentation before turning to AI tools to implement solutions.
A central theme in the discussion was Reed's provocative suggestion that "code quality doesn't matter as much as it used to" - a perspective he admits is "bizarre" for someone who built a career on code quality.
Reed contrasted his attitudes toward AI-generated content: "I would never have AI write a blog post, but I have no issue with AI writing my code." He explained this apparent contradiction by noting that code can be readily tested, verified, and corrected through established tools and processes, while written content maintains a distinctive voice that AI often fails to capture.
When Laporte asked about debugging AI-generated code that programmers might not understand, Reed described a "checkpointing" approach:
- Save functional versions of code
- If problems arise, don't try to debug - reset to the checkpoint and try again
- Rely heavily on tests - "I've had more tests in the last year than I've had my entire life"
Reed also highlighted a brilliant technique of asking AI to identify missing tests in a codebase and generate them automatically, leading to projects with "100% test coverage, every line of code is tested."
The conversation took a more serious turn when discussing AI's impact on tech employment. Reed acknowledged that many laid-off tech workers might not find similar roles again and admitted his own team is hiring fewer people than before.
When asked how he'd feel if his own job was eliminated, Reed responded with characteristic candor: "I've been trying to force myself out of a job for years. That's my dream. I just want to go and lay in a park and read books."
Despite his enthusiasm for these tools, Reed expressed concern about the broader societal impacts, noting that we are "underestimating the impacts of how this will work" especially when executives ask tech teams to essentially "dig their own grave" by implementing automation.
The conversation ended on an optimistic note about AI's potential to democratize programming. Reed shared a story about his brother, who has no programming background, creating a video game overnight using AI tools.
Jeff Jarvis asked what advice Reed would give to students interested in technology careers. Reed recommended experimentation and patience: "If I was going to talk to students today, I think I would just suggest just try it out. Try it on, see how it feels."
This sentiment resonated with Laporte, who expressed excitement about the potential for people to create despite lacking traditional programming skills: "You can achieve almost anything with enough patience."
