Persona: Senior Strategic Technology Architect & AI Research Fellow
Abstract:
This presentation, delivered by Anurag Mendhekar at the 2025 European Lisp Symposium, analyzes the trajectory of Common Lisp from its dominance in the "First Age of AI" to its current position in the era of Large Language Models (LLMs). Mendhekar posits that Lisp’s historical success was rooted in its perfect alignment with the "Epistemology + Heuristics" (Knowledge + Rules) paradigm of symbolic AI. However, the shift toward statistical AI—characterized by black-box neural networks and "glue" languages like Python—has marginalized Lisp’s traditional strengths.
The talk shifts to a critical analysis of the 2024–2025 software labor market, highlighting a dramatic collapse in engineering hiring as AI automation replaces "skilled coders." Mendhekar argues that the future belongs to "Deep System Visionaries" who understand entire stacks from first principles. He proposes a pedagogical revolution: teaching all computer science through Lisp. Because Lisp serves as a universal notation capable of "meta-dotting" (traversing abstractions) from high-level logic down to hardware, it is uniquely positioned to train the next generation of architects who must oversee AI-generated codebases.
Executive Summary: Lisp’s Role in the Age of Automated Programming
0:00 – Professional Lineage: Anurag Mendhekar details his background studying under Dan Friedman and working at Xerox PARC with Gregor Kiczales and John Lamping. His career spans the development of Aspect-Oriented Programming and the authorship of The Little Learner, a first-principles guide to deep learning using Scheme.
4:30 – The First Age of AI (Logic & Symbolic Reasoning): Lisp dominated early AI because it perfectly modeled the "Epistemology" (knowledge representation via cons cells/symbols) and "Heuristics" (computation via car/cdr/lambda). The REPL allowed for the interactive evolution of knowledge graphs.
10:32 – The 1987 Market Peak and Crash: By the mid-80s, the Lisp market was massive ($2.2B in 1987), fueling dedicated hardware (Lisp Machines) and high-performance computing (Thinking Machines). The "AI Winter" began when symbolic knowledge graphs became too complex and unmaintainable, leading to a total market collapse.
18:49 – The Shift to Statistical Intelligence: Modern AI moved from logic to statistics (Gaus/Bayes). Intelligence is now encoded in tensors and probability distributions rather than symbolic rules. In this "black box" era, Python became the dominant interface—not for its inherent intelligence, but as a convenient binding for high-performance C++/CUDA libraries.
29:00 – The Lisp Paradox (Nowhere and Everywhere): While Lisp usage in AI repositories is orders of magnitude lower than Python/C++, Lisp’s "heuristics" (garbage collection, recursion, functional programming, REPLs) have been absorbed into nearly every modern language (Java, JavaScript, Python, Ruby).
34:11 – The 2025 Labor Crisis: Mendhekar presents data showing a sharp decline in software engineering hiring starting in late 2022. As of 2025, the industry has reached net-zero growth. "Skilled coders" are being replaced by AI tools like Claude, Cursor, and Replit.
37:26 – The Rise of Lean Tech Giants: The industry is seeing "billion-dollar unicorns" with fewer than 50 employees. Startups no longer prioritize hiring "technology guys" but instead seek founders who can use AI to build applications directly.
40:08 – The Deep System Visionary: In a world where AI does the coding, humans must become system architects. These "visionaries" must understand the humanity of the product, the underlying algorithms, and the entire technical stack to debug and maintain AI-generated systems.
41:50 – Lisp as Universal Pedagogy: Mendhekar advocates for teaching all of Computer Science in Lisp. He argues Lisp is the only notation that permits "meta-dotting all the way to the hardware," allowing students to bypass the "arcania" of disparate language syntaxes and focus on first principles.
58:04 – The Irrelevance of Ecosystems: Mendhekar concludes that in an AI-driven world, the traditional barriers to using Lisp (lack of libraries/talent) are vanishing. AI can generate bindings and port code between languages instantly, leaving "designer bias" as the only remaining factor in language choice.
1:06:02 – Vibe Coding: The Q&A addresses "vibe coding" (Gen Z term for AI-assisted development). Mendhekar asserts that AI currently generates better code than many engineers and that the human role is shifting toward providing high-level rules and architectural guidance.
Key Takeaway: The era of the "skilled coder" is ending. To survive, the next generation must leverage Lisp’s unique power to model and understand systems from first principles, using AI as a high-speed construction engine while the human remains the "Deep System Visionary."