← Originals

Aman Sanger

3 interviews

What they argue

Aman Sanger is a thinker focused on the evolution of software engineering and the transition of Large Language Models (LLMs) from chatbots to agents. He explores the architectural shifts in development environments, the technical requirements for reliable agentic reasoning, and the move toward intent-driven programming.

  1. Intent-Driven Construction

    The evolution of software engineering will be defined by a shift from 'text editing' to 'intent-driven' construction. In this model, the editor acts as a bridge between human high-level logic and machine-generated implementation.

    From #447 – Cursor Team: Future of Programming with AI (episode)

  2. Platform Risk of AI

    Building AI products as extensions, such as VS Code plugins, creates a platform risk where the user experience is capped by what the host platform allows. This is a limitation because a competitor could build a better experience on a different platform.

    From Cursor.so: The AI-first Code Editor — with Aman Sanger of Anysphere (episode)

  3. Process-Supervised Reasoning

    To achieve reliable AI agents, we must move beyond 'outcome-only' evaluation and training toward 'process-supervised' reasoning. This involves breaking down complex problems into discrete, logical steps and labeling those steps to train the model to reason reliably.

    From ICLR 2024 — Best Papers & Talks (Benchmarks, Reasoning & Agents) — ft. Graham Neubig, Aman Sanger, Moritz Hardt) (episode)

    • A related idea Andrej Karpathy — Scaling and ShapingAI progress is driven by scaling up supervised learning (increasing parameters and data) and 'Shaping,' which is the professional practice of refining agent behaviors…Short Story on AI: A Cognitive Discontinuity.
    • A related idea Nathan Labenz — The Shift from Prediction to Task CompletionModern AI models are moving beyond simple next-token prediction toward being trained to achieve specific goals via reinforcement learning (RL). This shift means models…AI Scouting Report: the Good, Bad, & Weird @ the Law & AI Certificate Program, by LexLab, UC Law SF

  4. The Abstraction Stack Model

    To manage complex systems with AI, developers should use a mental model of navigating different levels of code. This includes viewing the system as a set of logical components and 'gestures' at a high-level/pseudocode level.

    From #447 – Cursor Team: Future of Programming with AI (episode)

    • A related idea Demis Hassabis — The AlphaZero ArchitectureTo build a system capable of complex reasoning, one should use a three-layer stack: a massive transformer for general knowledge (The Prior), a World Model, and a…Demis Hassabis — Scaling, superhuman AAIs, AlphaZero atop LLMs, AlphaFold
    • A related idea Simon Willison — The Three Pillars of CustomizationTo build successful AI applications, developers must leverage three distinct layers: the System API for stateful interactions, Dialogue, and other layers of…AGI is Being Achieved Incrementally (OpenAI DevDay w/ Simon Willison, Alex Volkov, Jim Fan, Raza Habib, Shreya Rajpal, Rahul Ligma, et al)

  5. Agentic Workflow for Code Review

    To move from 'writing' to 'delegating,' developers should use a framework for interacting with AI agents. This involves asking the agent to break a large feature into small, discrete units, such as writing a test function.

    From Cursor.so: The AI-first Code Editor — with Aman Sanger of Anysphere (episode)

    • A related idea Nathan Labenz — The Delegation FrameworkInstead of building rigid workflows, users should deploy agents by defining a clear end-state goal and identifying the necessary context. This shifts the focus from…AI in the AM — Week 1 Highlights (June 2026)

  6. Software Engineering (SWE) Benchmarks

    The frontier of LLM evaluation is moving from 'Write a Python function' to 'Fix this bug in a repository of 50 files.' Traditional benchmarks are considered too easy because they do not require the model to navigate complex, multi-file systems.

    From ICLR 2024 — Best Papers & Talks (Benchmarks, Reasoning & Agents) — ft. Graham Neubig, Aman Sanger, Moritz Hardt) (episode)

Every idea names the summary it came from. Written by a local model from this mind’s own work, never from memory.

Interviews 3

Open Aman Sanger in Originals