Neutral
The layoffs in the tech industry are real and applying for jobs is currently impossible.
The author observes the job market regardless of AI hype, stating layoffs are happening and applications are unsuccessful.
Source: I was wrong about AI coding agentsNeutral
Social media makes it hard to tell what is real about AI due to echo chambers.
The author says it's easier to get stuck in an echo chamber, making it difficult to discern truth.
Source: I was wrong about AI coding agentsAgree
AI tools are on a spectrum: some uses are slam dunk wins, but consistency decreases as complexity increases.
The author introduces a central thesis that AI effectiveness is not binary and must be evaluated per task.
Source: I was wrong about AI coding agentsDisagree
Claiming AI agents are completely useless is irresponsible.
The author explicitly states that after reconsideration, extreme negativity is irresponsible and untrue.
Source: I was wrong about AI coding agentsDisagree
Both the hype that coding is solved and the hate that AI agents are useless are wrong.
The author argues that neither extreme matches reality, setting the stage for a nuanced view.
Source: I was wrong about AI coding agentsNeutral
If coding were truly solved, much more complex products like a rewrite of After Effects would be consistently released, which is not happening.
Uses lack of high-complexity product releases as evidence against the 'coding is solved' claim.
Source: I was wrong about AI coding agentsAgree
More software products are being released and many people are creating products outside their usual capability, proving AI agents are not useless.
Presents increased shipping of software beyond individual skill as evidence that AI agents provide real value.
Source: I was wrong about AI coding agentsAgree
It would be ineffective and irresponsible to use AI to write production code that you do not understand.
The author warns that long-term issues and lack of understanding outweigh the benefits for production critical code.
Source: I was wrong about AI coding agentsAgree
It would be irresponsible not to use AI for rapid prototyping of throwaway work.
Argues that for disposable code, the speed gain from AI is so high that not using it is wasteful.
Source: I was wrong about AI coding agentsNeutral
There exists a 'critical thinking line' on the AI effectiveness spectrum where the cost of using AI begins to outweigh the long-term benefits.
The author defines a conceptual boundary to help developers decide when to stop using AI.
Source: I was wrong about AI coding agentsAgree
AI works consistently for code review with a low chance of unnecessary noise.
Based on personal experience, the author finds AI useful for catching mistakes in pull requests.
Source: I was wrong about AI coding agentsAgree
Refactoring and manual boring tasks like deleting unused code are worth attempting with AI, even though agents are less consistent at them.
Suggests using AI as a first pass for tedious work, because the potential time savings justify the inconsistency.
Source: I was wrong about AI coding agentsDisagree
LLMs are inconsistent at debugging by the nature of how they work, making debugging a poor use case.
Explains that the probabilistic nature of LLMs undermines reliability for root-cause analysis.
Source: I was wrong about AI coding agentsAgree
Delegating non-thinking tasks to an agent and using AI to choose what to think about preserves deep thinking rather than sacrificing it.
Cites Mitchell Hashimoto's workflow where agents handle routine tasks while the developer focuses on critical problems.
Source: I was wrong about AI coding agentsDisagree
If you let an agent handle the core problem while you watch YouTube, the AI will produce terrible code and you will feel unfulfilled.
Describes a misuse pattern where offloading critical thinking leads directly to bad outcomes.
Source: I was wrong about AI coding agentsAgree
Working on the critical parts yourself and delegating the rest to agents with notifications disabled allows more time in a flow state.
Presents Hashimoto's approach as an effective method: the developer retains control, interrupts only when ready, and focuses on what matters.
Source: I was wrong about AI coding agentsAgree
AI for research and discovery can reveal solutions you did not know existed and helps cover blind spots, especially because hallucinations can be thrown away at this stage.
Acknowledges inconsistency but argues the value of uncovering unknown alternatives outweighs the noise in early exploration.
Source: I was wrong about AI coding agentsDisagree
Using AI to generate a plan risks outsourcing critical thinking and may cost more time reviewing than thinking yourself.
Warns that letting a 'random next word guesser' propose the first solution can anchor thinking and review effort can exceed time saved.
Source: I was wrong about AI coding agentsAgree
When working on new or challenging designs, typing out code and playing with folder structures is the process by which most programmers figure out what to do.
Cites Dax Rad to emphasize that hands-on coding is still essential for thought and design, not just a mechanical step.
Source: I was wrong about AI coding agentsAgree
Understanding the domain, thinking about types and functions remains crucial for building scalable systems that work well, even when using AI.
Reinforces that deep domain knowledge is not replaced by AI; it is a prerequisite for producing quality software.
Source: I was wrong about AI coding agentsAgree
Using AI to write code is acceptable only when you already know the shape of the output, such as implementing an existing pattern or a well-understood problem.
Sets a clear precondition: AI code generation works when the developer can verify correctness against a known expectation.
Source: I was wrong about AI coding agentsAgree
To use AI for code generation effectively, you must care more about craftsmanship and keep your codebase a perfect example of what future code should look like.
Argues that high-quality, consistent codebases are essential for AI to generate correct code, raising the bar for developers.
Source: I was wrong about AI coding agentsAgree
Domain-driven design and hexagonal architecture with clear separation boundaries reduce the scope of changes and the risk of AI hallucinations leaking between components.
Recommends specific architectural patterns to contain AI-generated code and minimize side effects.
Source: I was wrong about AI coding agentsAgree
Unit tests are important because they help understand AI-provided solutions better and prevent regressions.
Frames testing as a safeguard and comprehension tool when incorporating AI-generated code.
Source: I was wrong about AI coding agentsAgree
Using AI for code is hit or miss; you must be quick to stop if it is not working because begging RNG to succeed is a waste of time.
Emphasizes the probabilistic nature of AI and advises cutting losses early to avoid investing time in inconsistent outcomes.
Source: I was wrong about AI coding agentsNeutral
Despite widespread AI adoption, complex projects are not seeing the promised productivity boost because AI does not make all tasks more efficient.
Observes a gap between demo-speed improvements and real-world complex project gains, explaining why extrapolation fails.
Source: I was wrong about AI coding agentsAgree
The trick to using AI effectively is understanding which tasks it does well and offloading low-priority tasks so you can focus on more important work.
Summarizes the core strategy: strategic delegation based on a realistic assessment of AI strengths.
Source: I was wrong about AI coding agents