Glean
Inside Glean's knowledge workflow: access, freshness, and answer review
What Glean documents about knowledge sources, document permissions, freshness, and answer feedback—and what an enterprise team should test.
Read the teardown →Make the next
AI decision clearer.
Deciding whether an internal-knowledge pilot can expand? Inspect the evidence, the failures, and the work still needed.
Research and practical reviews for enterprise product and platform leads.
Start with a worked pilot review →Four passing cases. One access failure.
A fictional knowledge pilot saves 13 active minutes across its passing cases. Expansion stays blocked. Follow the evidence into an editable decision packet.
Try the completed example →Read the research and method →Free · no account · fictional example clearly labeled
An answer has to earn
its way through.
Useful information is only the beginning. Source approval, customer access, and supporting evidence are separate decisions.
Read the Fin research →A visual analogy for proposed checks—not a simulation of Fin. Passing these checks alone does not establish production readiness.
Built with Anatomy ↗Read the model in plain text
An answer carriage moves along a rail. Each gate represents one independent condition. A closed gate stops it; opening one gate does not open the others.
- Source approved: The content owner has approved this material for customer answers.
- Customer access allowed: The information may be disclosed to this customer.
- Supporting evidence exists: The available source supports the proposed answer.
If a condition fails, stop the answer and define a human-review path.
Look beneath the interface.
The systems, decisions, and tradeoffs behind the product.
Intercom
Inside Fin's support workflow: knowledge, actions, and human handoff
How Fin connects support knowledge, external systems, testing, and escalation—and what a team should verify before letting it answer customers.
Read the research →GitHub
Inside GitHub Copilot's agent workflow
GitHub's Copilot stack connects retrieval, isolated agent execution, pull requests, policy controls, and measurement into one operating loop.
Read the research →From the Fin research
By Kyle Brierley
Sources checked September 19, 2026
Source selection is
a publishing decision.
Fin can answer from content that is not itself publicly visible, including internal articles and snippets. A document being absent from the public Help Center does not mean its contents are unavailable to customer-facing AI.
Intercom’s knowledge-source documentation ↗
Have a content owner approve the material used for customer answers. Test a question that should use an internal source and one that should never disclose internal information.
TTS’s proposed check, based on public documentation. Not a tested customer deployment.
Inspect the recorded evidence.
Dated company and product architecture snapshots. Check scope and sources before using a claim.
Bring a workflow.
Leave with a clearer plan.
A useful plan names the work, the owner, and the evidence still missing.
Open the Workbench →Start with work you recognize.
Research brief · 2026-09-15
Before choosing an agent platform, test what survives a failure
A research brief on session recovery, isolated execution, and the questions to bring to an agent-platform evaluation.
Browse all briefs →Take the research with you.
Read the explanation here. Give your assistant the same sources, qualifications, and evidence.
Founded by Kyle Brierley. Independent research with AI assistance.
Work through an AI stack decision with us →Follow the research
Get new research briefs and company teardowns by email, in your feed reader, or on X.
Email signup is hosted by Buttondown. RSS and X are available without an email subscription.