
Sign up to save your podcasts
Or


AI-levels:
https://ai-levels.org
Mailing list: (will send extra notes + info there next week)
https://gmail.us17.list-manage.com/subscribe?u=cf115741e983f14a4fd25e255&id=43d636b8d6
Prompt:
"# situation
A week after publishing I had this dream which suspiciously referenced emerging in a lift coming up from underground and arriving on what we thought were going to be level seven or eight which are the top two levels of AI levels and that being the ground floor level emerging so I think the dream is related to current and future expectations problems ideas issues opportunities missed ideas I haven't noticed or realised related to AI levels.
I can fill you in on some of the background thoughts and threats and concerns and ideas and opportunities etc all the uncertainties surrounding AI levels. But first I'd like your observations, opinions, etc.
# context
the full raw content of ai-levels.org is available to you directly and more cleanly in your second brain mcp at: /ai-levels/rubric.md - dont try to read the website
# task
Help me find what the dream is revealing in my subconscious thoughts or expectations or warnings or ideas or opportunities about AI levels."
You may be getting 5x improvements from AI - or bursts up to 20x - but how are other people reporting sustained 20x, 50x improvements in output?
... One aspect is this: do you only ever run your AI while you're present, or do you let it run unattended doing useful work while you're asleep?
Summary follows (Spotify bans us from pasting the full transcript)
Whether or not you personally identify with “vibe coding,” you’re probably, at this point, using some variation of agentic coding. It’s everywhere now. It’s in the IDE. It’s on the desktop. It’s in the cloud. It’s attached to repos. It’s quietly becoming the default way code gets written.The volume of code you’re writing now is conservatively five times the volume of code you were writing three years ago.I want to talk about the knock-on effects. The side effects. The bit I don’t think people are really discussing.
## Threaded Index (Narrative Map)
1. **AI coding isn’t just faster — it changes how many things you can attempt**
2. **Shepherded vs fully agentic coding: where most people actually live**
3. **The real productivity multipliers (5×, 10×, 20×) are already here**
4. **Why senior engineers and CTOs are coding again**
6. **The “graveyard curve” of unfinished AI projects**
7. **Psychological fallout: identity, shame, and organizational denial**
8. **Why code quality is no longer the bottleneck**
9. **The real bottleneck: everything around coding**
### 1. AI coding is already agentic — whether you call it that or not
Most people live in the middle: **shepherded coding**.
You watch the AI, accept or reject changes, course-correct, override sections, and sometimes just type the solution yourself because it’s faster than explaining the fix.
**Key point:** This range is fluid. Everyone moves back and forth across it.
### 2. Code volume has exploded — and that’s the baseline now
Most developers are writing **at least 5× more code** than they were three years ago.
If you’re only seeing a 10–20% improvement, you’re now in the minority — and probably not using AI effectively.
### 3. The multipliers are real — and they’re not hype
In my own work — commercial, personal, greenfield, and legacy — I regularly see:
* **5× overall** as a floor
* **10×** cases where days replace months
### 4. This has fundamentally changed who writes code
* CxOs and senior leaders who never coded
* Engineers who hadn’t written code in years
* Non-technical people writing production code
### 6. The part nobody is talking about: cognitive overload
When one person can run **5–6 projects in parallel**, a new problem emerges.
The emotional and psychological burden of:
* Constant context switching
* Multiple incompatible problem domains
* Different codebases, architectures, states of progress
…becomes overwhelming.
### 7. The “graveyard curve”
1. **Early phase**
One AI-powered project ships fast. The ROI feels magical.
2. **Acceleration phase**
You take on more projects.
3. **Warning phase**
You become the bottleneck.
4. **Collapse phase**
Months later:
* No projects are shipping
* Side projects are stalled
You now have:
* Six half-finished projects
* Ten more barely started
* No clarity on why any of them stopped
### 8. Why projects stall permanently
Each project hits a small speed bump.
Individually, manageable.
Collectively, impossible.
### 9. The psychological fallout
Repeated failure to deliver — even on self-promises — creates shame.
People hide:
* From colleagues
* From managers
* From themselves
And this is happening **everywhere**, largely unspoken.
### 10. Why optimizing code quality is the wrong fight
Since roughly ChatGPT-4, **code quality has not been the limiting factor**.
The real bottleneck is everything *around* coding.
### 11. What can we do? (No silver bullet)
There’s no universal answer.
But some directions are clear:
* Stop optimizing code generation
* Start optimizing **shipping capacity**
* Actively limit parallel projects
* Borrow from corporate innovation models that manage many experiments without overload
* Explore “AI operating system” approaches that use AI to manage the meta-work
From the publisher's feed