HS137: Did AI Turn “Everybody Codes” into “Nobody Codes”?
John Attil-Johnson and John Burke discuss how AI coding tools have fundamentally changed the "everybody codes" strategy, arguing that while AI can generate code quickly, logical thinking and code comprehension remain essential skills. They contend that the focus should shift from teaching everyone to code to ensuring everyone can read code and think logically to catch AI-generated errors.
Summary
The episode opens with listener follow-ups and friendly banter between the hosts about their long relationship before transitioning to the main topic. A detailed discussion about firewall definitions clarifies that while firewalls are theoretically configured as "deny all by default," in practice they often function differently, leading to broader discussion about zero trust models and defense-in-depth security strategies.
The main topic examines how AI coding has transformed the "everybody codes" movement that preceded it. John Burke notes that the push to get more people coding—particularly women—historically signals a field past its peak and about to be devalued, citing the example of Soviet doctors being predominantly female but low-status. The hosts discuss whether everyone should code given that AI can now generate functional code quickly.
John Attila-Johnson argues that coding's true value lies in teaching logical, detailed, precise thinking rather than syntax knowledge. He suggests that studying any discipline requiring logical precision—whether Pascal, Latin, or formal coding—provides benefits that AI cannot replace. Both hosts emphasize that AI is poor at consistency and logic, making human oversight critical.
The conversation explores technical debt, arguing that AI tends to incorporate poorly-written code into new programs because it lacks the logical discrimination to identify and exclude it. They discuss how the rapid improvement curve of AI tools makes revisiting code from weeks ago less attractive than recreating solutions fresh, potentially eliminating the debugging benefits of reusing mature code.
The hosts settle on a distinction between reading code and writing code: everyone should be able to read code with sufficient fluency to identify logical errors in AI output, but writing code fluently may no longer be necessary for most people. They acknowledge that computer science remains a vital field for fundamental research and mathematical discovery, but the pathway from coding bootcamps to lucrative careers has fundamentally changed.
About this episode
“Everybody codes” was an enterprise buzzword. In this era of AI vibe-coding and single-use coding, should everyone code? Should anyone code? John and Johna talk about enterprise strategies with respect to coding in the AI era, including what expertise to look for in employees. AdSpot Sponsor: Meter Meter delivers full-stack networking—wired, wireless, and cellular—to leading<a class="excerpt-read-more" href="https://packetpushers.net/podcasts/heavy-strategy/hs137-did-ai-turn-everybody-codes-into-nobody-codes/" title="ReadHS137: Did AI Turn “Everybody Codes” into “Nobody Codes”?">... Read more »</a><img height="1" src="https://feeds.packetpushers.net/link/22503/17370400.gif" width="1" />
Key Insights
- Burke observes that historical pushes to recruit women into fields often signal that field is past its peak and about to be devalued, citing Soviet doctors as an example where female-dominated professions had lower status and compensation.
- Attila-Johnson argues that coding's primary value is teaching logical thinking and precision rather than language syntax, suggesting Pascal or Latin could serve the same purpose and the skill matters more than the tool.
- The hosts contend that AI is fundamentally poor at logical consistency and reasoning, making human oversight essential regardless of how much AI improves, as logical flaws cannot be fixed by better models alone.
- Burke identifies a phase shift in code usage where rapid AI improvement makes recreating solutions from scratch faster than searching for or understanding existing code, changing incentives around code reuse and maintenance.
- The hosts argue that AI tends to incorporate poor-quality legacy code into new solutions because it lacks the logical discrimination to identify and exclude substandard implementations, thus amplifying technical debt.
- Attila-Johnson claims that the rapid rate of environmental change in software makes human comprehension of the deployment environment increasingly challenging, requiring machine tools like AI to navigate that complexity.
- Burke contends that flow diagrams generated by AI cannot be trusted to accurately reflect the actual code logic, as AI provides no guarantee the diagram aligns with the code it generates.
- Both hosts conclude that everyone should be able to read and understand code logic well enough to catch AI errors, but becoming proficient at writing code is no longer necessary for most people given AI capabilities.
Topics
Transcript
Hi, I'm John Attil-Johnson, CEO of Nemertes, here with my co-host. John Burke, CTO of Nemertes. And you're listening to Heavy Strategy, the show that tries to ask the right questions, not give the right answers. And before we dive into today's topic, we have a bunch of questions from listeners, so this is great. from listeners, so this is great. Before we dive into today's topic, just a quick word from our sponsor, Meter, which is the company building networks from the ground up. Meter delivers a complete networking stack, wired, wireless, and cellular in one integrated solution. With Meter, you get fast, secure, scalable connectivity without the burden of managing multiple providers or tools. Meter scales from branch…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Everything Feed - All Packet Pushers Pods
HN837: Agentic AI to Reduce MTTR
Edouard Douleroux discusses practical agentic AI implementation for reducing Mean Time To Recovery (MTTR) in network operations. He describes a deterministic system combining agents, fine-tuned language models, MCP servers, digital twins, and a YAML-based single source of truth to automate incident detection, diagnosis, and resolution testing.
IPB205: Moving to Disable IPv4
Frank Martin discusses the challenges and strategies for disabling IPv4 and deploying IPv6 in enterprise data centers, including operational hurdles, vendor coordination, and cultural change requirements. He introduces drafts on IPv6-only deployment and a retry-over-IPv6 mechanism to safely transition away from IPv4.
N4N061: Your First Wi-Fi Network Part 2
Holly Podbilak and Ethan Banks discuss advanced Wi-Fi technologies in Part 2 of their networking basics series, covering MIMO, OFDMA, MLO, and the evolution from Wi-Fi 4 through Wi-Fi 7, explaining how each technology addresses bandwidth and interference challenges in increasingly dense wireless environments.
TCG081: Network Automation Forum: From Simple Survey to Global Community
Network Automation Forum (NAF) emerged organically from a 2022 survey question about why network automation adoption lags in the industry, evolving into a global community-driven conference series with events across two continents. The founders emphasize that NAF's success stems from its commitment to practitioners-for-practitioners ethos, vendor neutrality, and willingness to listen to community feedback while protecting foundational values.
NAN128: Beyond RAG: Making Network AI Truly Agentic
Julia Santella, a Cisco solutions engineer with a software background, shares research findings on AI adoption in network automation, emphasizing that AI should augment rather than replace deterministic automation, and that critical skills for 2027 include learning to validate AI outputs rather than blindly trusting them.