Blog
Research notes on AI employees, internal workflows, governance.
Why guardrails are not enough for enterprise AI security
Policy guardrails do not stop a reasoning agent. Physical separation between the agent, critical systems, and approval logic is the architecture that actually works.
How Grail keeps secrets out of agent context
Grail now masks likely secrets before they reach the agent, preserves that protection across thread context, and keeps admins in control.
AI transformation is an operating model
Most AI transformation projects stall because they start with model demos instead of jobs that need to be redesigned.
AI employee vs AI workflow
A narrow workflow and a role-shaped AI employee are not the same thing, and buying them as if they are creates bad deployments.
The first 10 workflows worth automating with AI
The best early workflows are painful enough to matter and narrow enough to govern well.
Approvals are the product
The agents that survive production are the ones that know when to stop and make that stop legible.
B2B AI search is still search
AI answers change how content gets cited, but they do not remove the need for helpful pages, clean structure, and real authorship.
People-first content beats content factories
The temptation is to publish faster. The durable move is to publish clearer, narrower, and more useful pages.
What OAI-SearchBot, GPTBot, and ChatGPT-User actually do
If you treat every OpenAI bot the same, your robots policy is probably doing something you did not intend.
Building audit trails for AI employees
An audit trail is not a verbose log dump. It is the record that lets a human reconstruct what the agent was asked to do, what it did, and why.
Why exception queues decide whether AI workflows survive
If the exception queue is vague, overloaded, or ownerless, the workflow will look smart in a demo and exhausting in production.
What good AI search pages look like for B2B
Good B2B AI-search pages answer one real question clearly, show their source logic, and make the next step obvious.
Your first finance AI rollout should not start with a chatbot
The winning first move in finance is usually a narrow workflow with obvious controls, not a generic AI assistant sitting on top of everything.
Internal links are how AI systems understand your offer
One good page matters. A connected set of pages matters more because it teaches both humans and models how your offer actually fits together.
Agent testing is part of the product
If the team cannot predict how the workflow behaves in the ugly cases, the problem is not just model quality. The product is unfinished.
Why procurement is a better first AI workflow than most teams think
A lot of teams overlook procurement because it sounds unglamorous. That is exactly why it is often a better first workflow.
Business and IT should co-build AI agents
Treating AI rollout as either pure tooling or pure process is how companies end up with pilots nobody wants to own.
Employee lifecycle workflows are underrated AI rollouts
A lot of teams chase splashier workflows first, but employee lifecycle work often has the right mix of repetition, structure, and obvious control points.