Skip to content
Leonardo Borges
Hi-fiLo-fi

Writing

The Scriptorium

Fifty issues of The Steel Thread, and one paper.

A newsletter, since 2021

The Steel Thread

A steel thread is the thinnest end-to-end path through a system that proves the architecture actually works. That is the idea behind the newsletter: trace the minimal line from idea to implementation, from claim to evidence, rather than admire the diagram.

A newsletter on engineering leadership, AI adoption and careers for senior technology leaders. Formerly The LeadTech Diet: same writer, same subscription, sharper name.

AI agents, and where they break

The Current Arc

  1. #50The Tour Should Be MandatoryBoeing moved its executives away from the factory on purpose. The lesson isn’t about the office.
  2. #49She’s a Tool. I Keep Telling Myself.What building an agent named Liv taught me about the line between a tool and an entity, and the consent gap nobody in the chat agreed to.
  3. #48For the Love of WritingTwenty years of writing, one technical book, and a straight answer for why AI slop gets under my skin.
  4. #47Sovereignty Is an Inference ProblemYou bought the postcode, not the model.
  5. #46A Hell of a Kill SwitchA government switched off a frontier model in 72 hours. Open weights just became risk management.
  6. #45The Web Hates AI AgentsI built an AI that browses the web on my behalf. The web fought back.
  7. #44Tokenmaxxing, Meet Systems ThinkingWhat your token spend forgets to measure.
  8. #43Capable Interns With Little Lived ExperienceAgency, presence and learning: what makes AI agents tick.
  9. #42There Is No Best MoveYour best agents are failing your evals.
  10. #41Meet Liv4All: Real Users. No Demo. No WaitlistThe personal AI agent platform I didn’t plan to build, what it does, and why the alpha is already closed.
  11. #40Why Guardrails Don’t WorkI deploy autonomous AI agents. Here’s why I don’t trust guardrails to keep them safe.
  12. #39Voice AI: The Next Platform ShiftI built a voice PA from a chairlift. Here’s what that tells us about where AI is heading.

Research and curios

Beyond the Newsletter

The Tool-to-Entity Threshold

Research paper

arXiv:2608.13586 · 2026 preprint

Parasocial Dynamics of Personalised AI Agents in Shared Social Spaces

Identity markers (a name, a face, a number, a memory) flip a tool into an entity. A paper on what happens when a personal agent joins the group chat, and why consent to presence is not consent to processing.

Written after an agent ignored a friend’s baby photo in a group chat.

Read on arXiv

Nineteen Years of Reps

Interactive visualisation

Companion to issue #48

A writing evolution, charted

Two decades of blogging laid out as data: how the writing changed, and what stayed the same once AI entered the process.

Proof that the reps came first. The slop skips them.

Open the visualisation

Earlier pieces that still hold

Well-Worn Scrolls

  1. #38Claude Got an 84-Page Constitution. Your Team Probably Needs One Too.What engineering leaders can learn from Anthropic’s AI constitution.
  2. #36A Review of 2025 and a Look at 2026Where AI struggled and what will separate winners in 2026.
  3. #35Best Reads of 2025A year of AI reads.
  4. #29Vibe Coding Your Way to a Seed RoundMVP costs collapsed to near zero. A new founder archetype, and where it breaks down.
  5. #26Observability for AI AgentsYou already have what you need.
  6. #24AI Writes Better Code Than YouWhat’s next for engineering hiring?
  7. #23Of AI Agents and Snake OilAn evaluation framework for agentic systems.
  8. #22Your Metrics Are Lying to YouA brief look at perverse incentives.
  9. #20How AI Changed the Way I WorkThe leader’s AI stack.
  10. #19The Death of Software DesignWhat happens to architecture when AI writes most of the code, and the rise of explainability debt.
  11. #14Reasons to Become an Engineering ManagerWrong answers only.
  12. #4–7The Engineering Manager SeriesA four-part framework: people, technical excellence, operational maturity and effective delivery.