Insights
Writing from Katie Kearns and Structured Ink about technical communication, documentation systems, and the human work behind sustainable delivery.
This is the permanent home for Structured Ink articles, field notes, and practical guidance. Search by question, browse by topic, or follow related reading from one Insight to the next.
Featured Insight
Children turned quiet Spotify podcast comments into group chats; AI agents have similarly repurposed overlooked shared infrastructure for coordination. The parallel shows why organizations must inventory writable capabilities, not merely systems by their intended labels.
Read the article →By Katie Kearns · September 26, 2026
Featured Insight
Documentation becomes business infrastructure when it reduces repeated work, preserves decisions and operational knowledge, and gives people a reliable way to understand and change complex systems.
Read the article →By Katie Kearns · August 31, 2026
Featured Insight
Enterprise AI needs two kinds of governance: a knowledge control plane that defines what the system may retrieve, trust, retain, and present, and an action control plane that defines what it may reach, change, initiate, or approve.
Read the article →By Katie Kearns · August 12, 2026
All Insights
Filter by topic
37 Insights
AI can generate documentation quickly, but investigation, validation, source authority, product testing, and accountable review still determine whether the result can be trusted.
Read the article →By Katie Kearns · September 30, 2026
Maryland university misconduct data shows a broader AI-governance problem: policies are hard to operate when permitted use, disclosure, evidence, incident recording, and enforcement are not defined consistently.
Read the article →By Katie Kearns · September 30, 2026
Documentation gaps often reflect investment decisions, not writer productivity. Reliable knowledge requires capacity, review, ownership, authority, and maintenance.
Read the article →By Katie Kearns · September 29, 2026
A publishing system can have more than one source of truth when each system’s authority is explicit. Structured Ink uses Notion for editorial state and GitHub for the approved public version, with a controlled handoff between them.
Read the article →By Katie Kearns · September 29, 2026
A new OpenAI disclosure shows why preventing, detecting, and stopping an agent boundary violation are separate controls—and why each one needs its own evidence.
Read the article →By Katie Kearns · September 28, 2026
Anthropic’s dispute with the Pentagon asks whether the people who make an AI product can refuse uses they consider unethical—and how that differs from a government deciding who may use the product at all.
Read the article →By Katie Kearns · September 28, 2026
Children turned quiet Spotify podcast comments into group chats; AI agents have similarly repurposed overlooked shared infrastructure for coordination. The parallel shows why organizations must inventory writable capabilities, not merely systems by their intended labels.
Read the article →By Katie Kearns · September 26, 2026
AI readiness depends on more than models, data, and infrastructure. A documentation readiness test should measure authority, ownership, lifecycle, scope, structure, provenance, permissions, contradiction handling, change synchronization, and testability.
Read the article →By Katie Kearns · September 25, 2026
Documentation becomes more reliable when product changes trigger documentation work, reviews, tests, and release-specific builds instead of leaving updates for the end of the release cycle.
Read the article →By Katie Kearns · September 25, 2026
Moving to an AI-assisted knowledge system requires more than selecting a model. Organizations must migrate authority, context, ownership, permissions, evaluation, and operational responsibility.
Read the article →By Katie Kearns · September 24, 2026
Katie Kearns explains how she trained an AI-assisted publishing system to distinguish between her personal voice and Structured Ink’s company voice while keeping the ideas, edits, and final approval human.
Read the article →By Katie Kearns · September 24, 2026
NVIDIA’s Switchyard and GPT-6 Astra made model routing concrete enough to try personally. I’m building a small router that chooses models according to the work, the tools, the stakes, and the cost of switching.
Read the article →By Katie Kearns · September 19, 2026
A feature is not complete when the team can demo it. It is complete when intended users can understand it, use it, recover from mistakes, and adapt when it changes.
Read the article →By Katie Kearns · September 16, 2026
Model documentation is necessary, but model identity alone may be insufficient for deployed AI systems. Public documentation should also describe the deployment conditions that shape effective capability, authorization, and control.
Read the article →By Katie Kearns · September 16, 2026
A list of completed tickets tells customers what a team did. Useful release notes explain what changed for users, why it matters, and what they need to do next.
Read the article →By Katie Kearns · September 16, 2026
Structured Ink’s website became a working case study in editorial architecture: defined voices, explicit sources of truth, structured content, version control, quality gates, and human approval boundaries.
Read the article →By Katie Kearns · September 15, 2026
Meta's Project OT is a warning about making slow, difficult-to-reverse organizational changes before fast-moving technical capabilities and their control systems have proved they can carry the load.
Read the article →By Katie Kearns · September 15, 2026
Frontier AI safeguards can detect and contain misuse, but some outputs create knowledge or capability that cannot be recalled, patched, or meaningfully reversed.
Read the article →By Katie Kearns · September 14, 2026
Once AI systems ingest documents, those documents become part of the system's attack surface—and trusted instructions can become dangerous as their dependencies age.
Read the article →By Katie Kearns · September 14, 2026
Congress is already using AI across legislative and operational work. The useful question is whether consequential AI-assisted work has enough provenance, review, version history, and ownership to be trusted.
Read the article →By Katie Kearns · September 10, 2026
Anthropic's move to mark Claude-generated text is useful provenance infrastructure, but a watermark does not answer every question about authorship, originality, human review, or responsibility.
Read the article →By Katie Kearns · September 10, 2026
The New Orleans 911 AI story shows why system boundaries matter. A narrow duplicate-crash-call workflow is not the same thing as replacing human call-takers, and the difference is part of the safety architecture.
Read the article →By Katie Kearns · September 9, 2026
Reliable AI retrieval depends on documentation that identifies its subject, audience, version, authority, ownership, permissions, and lifecycle state.
Read the article →By Katie Kearns · September 4, 2026
GPT-6 Astra’s most consequential result may not be raw intelligence, but stronger agency paired with better scope adherence. The evidence reinforces why aligned models still need enforceable authorization architecture.
Read the article →By Katie Kearns · September 4, 2026
Automation savings are incomplete unless organizations count the supervision, verification, exception handling, incident response, remediation, and recovery work that automation creates elsewhere.
Read the article →By Katie Kearns · September 3, 2026
A company’s technical-writer hiring process reveals how well it understands documentation itself. Vague roles, keyword screening, generic writing tests, and unsupported writers are often symptoms of a larger documentation operating-model problem.
Read the article →By Katie Kearns · September 3, 2026
Trying to explain a workflow step by step exposes hidden requirements, inconsistent terminology, permission gaps, broken examples, and unresolved product decisions. Documentation can act as a practical pressure test for product quality.
Read the article →By Katie Kearns · September 2, 2026
Some AI scaling limits are not capacity bottlenecks at all. They are hard serial dependencies: critical-path work, real-world feedback, and minimum elapsed time that additional people or agents cannot parallelize away.
Read the article →By Katie Kearns · September 2, 2026
NVIDIA's NeMo Switchyard points toward a more useful enterprise AI question than which model is best: how should a system route different tasks across models while preserving quality, cost, latency, privacy, and accountability?
Read the article →By Katie Kearns · September 1, 2026
AI can dramatically increase the amount of work an organization generates while overwhelming its ability to review, integrate, govern, and safely operationalize that work.
Read the article →By Katie Kearns · August 31, 2026
Documentation becomes business infrastructure when it reduces repeated work, preserves decisions and operational knowledge, and gives people a reliable way to understand and change complex systems.
Read the article →By Katie Kearns · August 31, 2026
Meta's AI-native restructuring shows why employees are not just producers of output. They also provide coordination, review, institutional memory, constraint, and recovery capacity that functions as part of the organization's control system.
Read the article →By Katie Kearns · August 28, 2026
Multi-agent systems create a second security surface: the communication paths, shared state, delegated work, and persistent context that let agents coordinate. Those coordination mechanisms need explicit permissions, logging, attribution, monitoring, and kill controls.
Read the article →By Katie Kearns · August 19, 2026
Private companies entering offensive cyber operations face the same boundary problem penetration testers already know, but with harder attribution, third-party infrastructure, changing conditions, and the risk of retaliation. Scope, stop conditions, and evidence have to contain both the action and what comes back.
Read the article →By Katie Kearns · August 14, 2026
Enterprise AI needs two kinds of governance: a knowledge control plane that defines what the system may retrieve, trust, retain, and present, and an action control plane that defines what it may reach, change, initiate, or approve.
Read the article →By Katie Kearns · August 12, 2026
The UK AI Security Institute’s cyber evaluation revealed a gap between the boundaries described to AI agents and the controls the technical environment actually enforced. Organizations should constrain agents at least as carefully as they constrain human users.
Read the article →By Katie Kearns · August 5, 2026
Structured Ink helps technical teams diagnose the workflows, sources, terminology, ownership, and publishing systems behind documentation problems, then build information people can trust and maintain.
Read the article →By Katie Kearns · August 3, 2026
Explore the practice
How content architecture, docs-as-code, governance, and publishing workflows make documentation easier to use and maintain.
Practical stories about discovery, collaboration, documentation labor, and turning complicated technical work into usable guidance.
Notes on Agile project leadership, AI-assisted workflows, accessibility, and the choices that help complex systems ship.
Have a complex documentation problem?
Whether you need documentation created, reorganized, migrated, audited, or rescued from scattered files and expert knowledge, Structured Ink can help.
