New releaseThe Anthropic Book ↗

19 chapters  ·  corpus-verified  ·  audio narration

CONTEXT JAMMING

Field notes from inside the context window.

Hire me

Bret Kerr // Content design

Senior Content Designer & Content Strategist: AI content systems, technical clarity, editorial QA.

Ten years turning cybersecurity complexity into plain language at Mimecast. Now I write the instructions, knowledge sources and quality checks that AI agents work from, and I edit what they produce.

SignalThesisTestShip
Section 01Capabilities

What I do for content teams.

Six things a content team can hand me. Each card names the resume competency it answers and links to live work.
  • Content design for AI agents and skills

    Resume match · AI agents and reusable AI-powered workflows

    I write what an agent works from as a document with fixed parts: when to apply it, the inputs it accepts, the output contract, the voice and style rules, and the checks the output must pass.

    Read the skill files on the Claude Code skills page

  • Editorial QA of AI output

    Resume match · Editorial judgment on AI-generated content

    Two models research the same question from opposed briefs. I cite-check and reconcile the reports before drafting. Quotes are verified and numbers are recomputed, because two models agreeing is not evidence.

    See the recorded reasoning behind one published piece

  • Information architecture and terminology

    Resume match · Content strategy, information architecture and terminology

    I name things, group them, and write the labels people navigate by. On this site the menu is tested: every label must point to a real page, and each page sits under one menu item.

    Browse the Founder Files taxonomy

  • Technical simplification

    Resume match · Simplify complex technical concepts

    Ten years of making security products and threats clear for executives, sellers and customers. Today I rebuild research papers as interactive explainers with a glossary, stated limits and source notes.

    Open the interactive research explainers

  • Research synthesis into narrative

    Resume match · Content strategy, information architecture and terminology

    I turn a large, mixed set of sources into a narrative a leadership team can act on: a point of view, the category language, and the evidence behind each claim.

    See the executive narrative work at ACRA Insight

  • Multi-format content systems

    Resume match · Content strategy, information architecture and terminology

    One source, many formats. The Anthropic Book ships as web chapters, PDF, EPUB and audiobook, and articles on this site carry a clean Markdown copy for AI systems.

    Read The Anthropic Book

Section 02Case studies

Six case studies, all live.

Each one follows the same order: the problem, my role, what I made, the result, and where to see it. Results are limited to what the linked page states.

Content design for AI

Skills: instructions an AI agent follows the same way twice

  • Agent instructions
  • Prompt engineering
  • Evaluation
  • Content specs
  • Markdown
  • MCP
Problem
An agent does what its instructions say. A loose prompt gives a different answer on every run, and nobody can review a prompt that lives in one person’s chat history.
My role
Content designer and builder. I wrote each skill, tested it and published it.
What I made
  • Skills written as documents with a fixed structure: when to apply, inputs accepted, output contract, voice and style, and a verification gate the output must pass.
  • A two-stage research pipeline. The first skill turns a brief into a structured research prompt. The second turns the finished research into a longform article, an infographic spec and social copy.
  • Skill Memory Bank, which compiles a working conversation into a SKILL.md package that loads in Claude Code, Cursor, Copilot, Codex and Gemini CLI.
Result
  • Best in Show, Lane 2, at the Open Accelerator Hackathon (Red Hat, IBM and the Mass AI Hub), Boston, June 27, 2026.
  • In Skill Memory Bank’s evaluation harness, retention on a fixed input went from 0 of 3 without the skill to 3 of 3 with it.
  • The open-source skills are published under Apache 2.0.

Editorial QA · LLM output evaluation

A two-model research loop with an editor in the middle

  • LLM output evaluation
  • Editorial judgment
  • Source checking
  • Research synthesis
  • Agentic systems
Problem
A model’s research report reads as confident whether or not it is right. Publishing one unchecked puts its errors under your name.
My role
Editor and workflow designer. I write the two briefs, check the sources and decide what the piece claims.
What I made
  • Two opposed deep-research briefs: Gemini maps the territory and GPT argues against the thesis.
  • A reconciliation step that cite-checks both reports and records each conflict before any drafting.
  • A public record of one run: Claude Sonnet 5’s full reasoning, paired step by step with the finished dashboard it produced.
Result
  • In the recorded run, 45 thought units across 9 phases reconciled 2 reports, and 7 corrections were adopted from the adversarial report.
  • The run overruled the report that agreed with me and adopted the adversary’s correction before any publishable copy was written.

IA · terminology · GEO and AEO

Information architecture, taxonomy and findability

  • Information architecture
  • Taxonomy
  • Terminology
  • Navigation labels
  • GEO and AEO
  • Structured data
Problem
A site of about 300 pages with one author drifts. Menus overlap, labels stop matching pages, and AI answer engines cannot tell what a page is about.
My role
Content strategist and information architect. I own the navigation, the labels, the taxonomy and the machine-readable layer.
What I made
  • Founder Files: 77 serialized profiles classified on the same six dimensions, with six named career shapes, filters by lab, cluster and archetype, and a plain note on how matches are counted.
  • One navigation source for the header, the footer and the breadcrumbs. Two overlapping menus, Hire and Work, became one.
  • A jump menu on the homepage that separates the hiring surface from the publication.
  • A layer for answer engines: llms.txt, a Markdown copy of each article, structured data, and a guide to GEO with its own defined metric vocabulary.
Result
The menu rules run as automated tests. A label that points to a missing page, or a page listed under two menu items, fails the test run.

Technical simplification

Interactive explainers: research papers made usable

  • Technical simplification
  • Plain language
  • Glossaries
  • Instructional sequencing
  • Captions
Problem
AI research papers are written for other researchers. People in marketing, sales and leadership need the finding, its limits and its vocabulary, and a PDF gives them none of that quickly.
My role
Writer, designer and builder.
What I made
  • 19 sources rebuilt as interactive explainers, from neural scaling laws to Constitutional Classifiers.
  • A repeatable page structure: numbered sections, a glossary, a ledger of what is measured, modeled and still unknown, and source notes.
  • Transformer Atlas, where the reader runs an experiment on GPT-2 small and the page states its own limits.
  • Short explainer films with burned-in captions for sound-off viewing.
Result
In Transformer Atlas, every number is recomputed from the model’s real residual stream.

Content strategy · structured long-form

The Anthropic Book: structured long-form from 727 sources

  • Content strategy
  • Structured content
  • Source verification
  • Multi-format publishing
Problem
The source material on Anthropic’s first five years ran to 727 documents and about 1.4 million words. No reader can hold that, and neither can a single model context.
My role
Author, editor and system designer.
What I made
  • A book with a prologue, numbered chapters, an interlude, and a coda that explains the method.
  • One rule for quotes: every quoted passage is verbatim from the corpus and substring-verified before publication.
  • Four formats from one source: web chapters, PDF, EPUB and audiobook.
  • A public write-up of the tech stack and pipeline behind it.
Result
Published and live: 727 source documents, 21,222 retrieval chunks, and 1 hour 41 minutes of narration.

Research synthesis · executive storytelling · Dec 2025 to Apr 2026

Executive narrative for an enterprise cybersecurity client

  • Research synthesis
  • Executive storytelling
  • Messaging
  • Competitive narrative
  • Content systems
Problem
An ambiguous, AI-era security market had to become executive narrative, campaign architecture and reusable content systems within one quarter.
My role
The one senior operator, working with two frontier models. Editorial judgment controlled the final argument, language, evidence and release decision.
What I made
  • Six strategic deliverable clusters, published as Exhibits A to F: an alliance thesis, a competitive campaign, C-suite positioning, an innovation center narrative, an AI-to-Drupal web workflow, and field reportage.
  • A seven-step narrative model that runs from research corpus to editorial QA.
  • A 104-page competitive campaign architecture.
Result
  • All six clusters shipped in one quarter.
  • The alliance thesis was adopted as a C-suite strategic priority.

Figures are as stated on the ACRA Insight page. Proofpoint is referenced there as a past client engagement context; this page is not affiliated with or endorsed by Proofpoint. Client-sensitive sources, review history and implementation detail are withheld.

Section 03Mimecast · 2015 to 2025

Before AI, the same job.

Ten years at Mimecast, doing content design with a camera: guidance for presenters, step-by-step product help, onboarding, and the content plan for whole events. Internal work is described here and not shown.
  • Guidance writing2021

    Presenter guidance for a virtual product event

    A two-page guide that gets non-professional presenters to record a usable session from home. It covers the slide template, the room, a two-minute test recording with pass criteria, and the exact recording settings with screenshots. Every rule comes with its reason.

    • About 550 words
    • Three screenshots
    • Reused as a general template

    Shows · Guidance writing
    Sample available on request

  • Product how-to2021

    A product how-to series: configure, then troubleshoot

    Two parallel video tracks for one product feature, three steps each. Each step covers a single task in two to five minutes, and each went through three versions in one week.

    • Six videos
    • Two three-step tracks

    Shows · Instructional sequencing
    Sample available on request

  • Onboarding2020

    API onboarding in three steps

    Overview, then onboarding, then generating API keys. Three videos of about two and a half minutes each, made for a virtual summit.

    • Three videos
    • 2:21, 2:23 and 2:40

    Shows · Onboarding
    Sample available on request

  • Virtual events2020

    Content architecture for a virtual summit

    A numbered run of show: an opening reel, eight product tip sessions, breakout sessions, keynotes, a pre-show countdown and a regional re-cut. One session template kit held it together, with open and close animation, lower thirds, and layouts for one or two presenters.

    • Eight tip sessions
    • One template kit
    • Regional re-cut

    Shows · Event content architecture
    Sample available on request

  • Videos2021

    Short security explainers for social

    A short-form series built on one template, with question-and-answer scripts and final caption files. Episodes covered ransomware and securing the hybrid workforce.

    • Scripts of about 400 to 500 words
    • Captioned

    Shows · Technical simplification
    Sample available on request

  • Webinars2021

    A research report turned into a webinar series

    A published brand-protection report became a captioned webinar series, alongside regional webinars and a three-part series on privacy law.

    • Captioned
    • Regional editions

    Shows · Technical simplification
    Sample available on request

  • Customer stories

    Customer stories turned into sales clips

    A transcript- and GPT-assisted workflow. Webinars and customer interviews became clips, quotes and campaign assets, packaged in Seismic for sales.

    See published people and customer stories on the video production page

    Shows · Customer research and synthesis
    Internal work · described, not shown

  • Sales enablement2019 to 2020

    A news show for the sales team

    Internal episodes built from 20 or more numbered segments, each with a script: product, competitive and regional updates of one to three minutes each.

    Shows · Enablement
    Internal work · described, not shown

  • Executive communications

    Executive communications

    Direct creative partner to the founder and CEO through the public-company years and the take-private. Leadership messaging and large internal shows such as sales kickoffs. Internal work, so it is not shown publicly.

    Shows · Executive storytelling
    Internal work · described, not shown

Section 04Tools

Tools I work in.

AI tools for research, drafting and evaluation. Developer tools, because the content ships as a site. Media tools from ten years in a studio.
AI
  • Claude
  • Claude Code
  • Cursor
  • Codex
  • MCP
  • Gemini
  • ChatGPT
Build and publish
  • Next.js
  • TypeScript
  • Tailwind
  • Cloudflare
  • GitHub
  • Markdown
Media
  • Adobe Premiere
  • After Effects
  • Photoshop
Next contract

Need a content designer who can also build the system?
Send me the brief.

Contract or temp-to-hire. Remote, hybrid in Boston and Cambridge, or on-site along Route 128 and 495.

bret.kerr@gmail.com · linkedin.com/in/bretkerr · contextjamming.com/contentdesign