What my dad taught me about AI coding in the 90s
Lessons from the 90s: Blindfold Chess and the Future of AI Coding
Originally inspired by a reflection on Askmike.org, posted August 23, 2026.
A Memory of the Late 90s
During a summer break in the late 1990s, my father and I spent time playing chess. While he had already taught me the fundamentals, one particular session stood out because of a strange twist in the rules.
Instead of sitting across from me, my dad stood up and moved to a chair on the opposite side of the room, facing away from the board. The game proceeded as follows:
- I would make a move on the physical board.
- I would announce the move aloud:
pawn from E2 to E4. - He would respond verbally with his move.
- I would then execute his move on the board to keep track of the state.
I recall a moment of confusion where I tried to move a piece from a specific square, only to realize the piece wasn't actually there. This highlighted the difference between seeing and understanding.
The Secret of the Blindfold Player
Contrary to popular belief, elite blindfold players do not possess a photographic memory of the board. Instead, they maintain a dynamic map of structured relationships.
"Their concentration is directed more toward relationships and legal possibilities than toward literally 'seeing' the board." — Claude Fable 5
They track specific variables:
- Piece Positioning: Where the critical assets are located.
- Zone Control: Which squares are currently under attack or defended.
- Geometry: The state of open files, diagonals, and pawn chains.
- Tactical Hooks: The presence of pins, forks, or "hanging" pieces.
- Delta Tracking: Exactly what changed after the most recent move.
Essentially, their expertise is a synthesis of:
From the Chessboard to the Terminal
While I never became a chess master, I found my passion in computer science and programming. Today, the landscape for software engineers is shifting violently due to the emergence of Large Language Models (LLMs).
We are currently in a transitional phase where everyone is searching for the "correct" way to integrate AI into their workflow. There is a dangerous trend emerging: "Vibecoding."
Vibecoding occurs when a developer:
- Spends minimal effort defining the requirements.
- Spends zero time auditing the generated code.
- Relies entirely on the "vibe" of the output.
The result? Software that is fragile, unsustainable, or completely broken.
The Evolution of Abstraction
The shift toward AI coding mirrors the transition to high-level languages, but with a critical difference.
| Era | Requirement | Abstraction Level | Focus |
|---|---|---|---|
| Low-Level | Deep hardware knowledge | Low | Pointers, Memory, CPU cycles |
| High-Level | Language-specific logic | Medium | Java/Python syntax, Frameworks |
| AI-Driven | High | Intent, Architecture, Review |
In the past, moving from Assembly to Python meant we could ignore memory management, but we still had to master a fixed set of logical rules. With AI, there is a tempting illusion that no knowledge is required at all.
The Paradox of Attention
Programming with AI is, in many ways, the inverse of blindfold chess. In chess, you must pay attention to every turn to survive. With AI, the tool is so capable that it invites laziness. Even an experienced developer using Claude Code might stop reading the output, letting the AI take the wheel entirely.
However, I argue that the exact same skills required for blindfold chess are what will separate the great AI-augmented programmers from the mediocre ones.
The "Mental Model" Advantage
The most successful programmers in the AI era are those who can maintain a deep internal representation of how code weaves together without needing to physically write every line. This involves:
- Viewing code as an expression of an abstract goal.
- The ability to simplify an implementation through deep thought alone.
- Balancing the granularity of communication (knowing when to be vague and when to be precise).
For me, building a mental model of code I didn't write is still a challenge. Finding the equilibrium between "talking in detail" and "talking at a high level" is an ongoing struggle.
Final Thoughts
As LLMs evolve, those who can "tame the beast" while remaining in control will operate at speeds previously unimaginable. They won't be limited by their typing speed or their ability to read boilerplate, but by the clarity of their mental models.
The ultimate irony? In chess, skill is easily ranked by ELO. In AI programming, quantifying skill will become nearly impossible, as the AI can be used to optimize any measurable metric we try to use for evaluation.