Am I holding this right? • Daniel Terhorst-North • GOTO Copenhagen 2026

Am I holding this right? • Daniel Terhorst-North • GOTO Copenhagen 2026

Title history
  • Am I holding this right? • Daniel Terhorst-North • GOTO Copenhagen 2026 — observed on 20261105-1421 UTC (YouTube)

by Daniel Terhorst-North · View the channel on PSI

Applied Sciences & Engineering Computing & Cybersecurity UMComputer programmingUMZSoftware Engineering
🎙 Daniel Terhorst-North 👥 1.1M 📅 November 5, 2026 ⏱ 42 min 👁 56 📄 expert opinion 🧭 2026-11-05
Available in: English (current) Français

Keywords

LLMClaudeadjacent engineeringTDDtheory of constraints

Summary

Daniel Terhorst-North shares his personal journey of using generative AI (specifically Claude) not to accelerate coding but to broaden his capabilities by offloading side concerns. He introduces the concept of ‘adjacent engineering’—using AI for tasks where he is competent but not expert, such as converting markdown to slides, building presentation tools, and analyzing his teaching materials. He also describes ‘adjacent discovery’ when such side projects become significant. He critiques the hype around LLMs, referencing Emily Bender’s ‘stochastic parrot’ metaphor, and emphasizes that AI excels at synthesis and pattern matching but lacks true understanding and self-regulation. He shares practical observations: LLMs are great at cleaning messy data and generating idiomatic code, but they are unreliable with tools and prone to errors like deleting domains due to missing guard clauses. He uses anecdotes and metaphors (e.g., Claude as an enthusiastic but careless puppy) to illustrate both potential and pitfalls. The talk concludes with advice to treat LLMs as powerful but fallible assistants, and to always verify their outputs.

166 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk’s value lies in its hands-on, pragmatic perspective from an experienced software developer. Terhorst-North grounds his arguments in concrete examples (e.g., converting JQ scripts, building a presentation tool) rather than abstract theory. His argumentation is coherent and persuasive, though largely based on personal experience rather than systematic data. He acknowledges the limitations of LLMs while highlighting their utility, offering a balanced view. The use of metaphors like ‘scrapheap programming’ and ‘stochastic parrot’ effectively communicate complex ideas. However, the lack of empirical evidence or comparative analysis weakens the scientific rigor.

Scientific Rigor, Source Quality, Title Accuracy

Terhorst-North references several well-known sources, including Emily Bender’s ‘stochastic parrot’ paper, James Noble and Justin Biddle’s ‘postmodern programming,’ and Goldratt’s ‘Theory of Constraints.’ These lend credibility, but they are cited in passing without deep engagement. The talk does not provide a formal literature review or methodology. The title ‘Am I holding this right?’ is a playful but apt question that frames the talk as a self-reflection on his use of AI, which is consistent with the content. There are no comments provided to analyze.

189 words

Title / Content Match

The title is a playful rhetorical question that accurately reflects the talk's theme of evaluating one's approach to using LLMs.

Quality & Reliability

6/10

The talk is a personal, experience-based account of using LLMs, with references to known concepts but no rigorous methodology. It offers practical insights but relies heavily on anecdote.

Chapters

Cited Sources

  • Daniel Terhorst-North's blog — Speaker's personal blog with articles on software practices and AI.
  • GOTO Copenhagen 2026 session page — Detailed abstract and slides download for this talk.
  • GitHub profile of tastapod — Repository of tools built and mentioned during the talk.
  • LinkedIn profile of Daniel Terhorst-North — Professional profile and additional resources.
  • GOTO conferences official page — Main website for GOTO events and content.

External References

Contribution & Novelties

The talk contributes a nuanced, practitioner-level framework for thinking about LLM use: ‘adjacent engineering’ (offloading side tasks) and ‘adjacent discovery’ (when side quests become products). It also challenges the ‘speed of code production’ narrative by emphasizing business impact and maintainability. The ‘scrapheap programming as a service’ metaphor is original and captures the reality that many LLM outputs are recombinations of existing code. Practical pitfalls, such as lacking self-regulation and the sunk cost fallacy, are valuable for users.

Pour aller plus loin :

  • Stochastic parrot (Wikipedia) — Origin of the term in relation to LLM critique.
  • Theory of constraints (Wikipedia) — Goldratt’s management paradigm referenced in the talk.
  • Training from the back of the room (Wikipedia) — A teaching methodology mentioned by the speaker.
  • Postmodern programming article (PDF) — The paper by Noble and Biddle that discusses code reuse and patterns.

140 words

Radar Profile

The radar profile shows a moderate quantity of information, lower technical depth, and moderate reliability, indicating a talk that is rich in practical anecdotes but not deeply technical or rigorously sourced. This aligns with a practitioner's perspective aimed at sharing experience rather than presenting formal research.

Reliability 5/10