Good research.
Ready for your agent.

Find the right article. Read the evidence. Keep the source attached. Our public research is available as clean Markdown and a structured index, alongside the pages you already read.

Read llms.txt ↗Open JSON index ↗

Public access · No API key · Generated with every site build

01 / DISCOVER

A map of the research

llms.txt groups our research and guides into a readable directory. It follows an emerging convention; support varies between agents.

02 / SELECT

Choose before fetching

The JSON index includes descriptions, content types, citation URLs, dates where known, author and editor profile links, suggested citations, retrieval fallback guidance, publisher information, byte sizes, and content hashes.

03 / READ

Content with context

Markdown preserves the page’s evidence, links, tables, and qualifications. Each export identifies the original page to cite.

Explore the actual index.

Every article is linked in this page’s HTML. Search and previews enhance the same public catalog your agent can use.

25 public pages. All entries are available below; filters work when JavaScript is enabled.

authors · 4697 bytes

Kyle Brierley — Founder & Author

Kyle Brierley is the founder of TheTechStack and a product leader with experience at Oracle and Tealium. His work spans enterprise AI workflows, customer data platforms, and API integrations. He writes about the decisions teams face when connecting AI to real business processes, including evaluation, permissions, and operating ownership.

View Markdown

teardowns · 10063 bytes

What Linear actually runs

The most opinionated product in project management runs a deliberately boring stack — TypeScript, Postgres, GCP — except for the one thing they built before almost anything else: the sync engine.

By Kyle Brierley

View Markdown

Discover → retrieve → cite.

Ask your assistant to use the index to find the completed AI pilot decision example, open its original article, and explain why rollout paused. If your tool cannot open that format, try its Markdown export. A successful answer identifies the access-control failure, keeps the fictional-example label, and cites the original page.

Use https://www.thetechstack.com/agents/index.json.
Find the completed AI pilot decision example.
Open canonical_url and explain the rollout decision.
If your tool cannot open it, try markdown_url.
Preserve the fictional-example caveat.
Cite canonical_url and link the credited author_url.
Use suggested_citation after checking the source.
If both formats fail, report the tool limitation instead of guessing.

Read the source example · Inspect the exported Markdown

The explorer demonstrates discovery and retrieval. It does not call a model or score an assistant’s answer.

What you can rely on.

Research and editorial resources identify Kyle Brierley where attributed. Stack records name the responsible editor and preserve credit to their original sources. The catalog also includes Kyle’s profile.

One published source

Exports come from the built page’s main content. They retain source links and fictional-example disclosures. Drafts, design studies, and interactive form state are excluded.

Honest freshness

Publication and update dates are included where the source specifies them. Unknown dates are null. SHA-256 hashes identify changed Markdown; they do not imply a new editorial review.

Stable entry points

Index schema version 1.0 provides canonical_url and markdown_url. Use those links instead of guessing paths. Cite the canonical HTML page.

Clear limits

This release uses explicit Markdown URLs. It does not provide an MCP server, a search API, or content negotiation. Making content accessible does not guarantee indexing, citations, or ranking.

Our next questions.

  1. V1 · Available here

    Markdown exports, a text directory, JSON catalog, and this retrieval explorer. Automated checks protect coverage, links, and evidence preservation.

  2. Next · Measure retrieval quality

    Publish a small, versioned benchmark: can assistants find the right source, retain caveats, and cite it correctly? Report failures alongside successes.

  3. Then · Improve discovery with evidence

    Study retrieval requests, stale links, and useful reader referrals. Add topic indexes, change feeds, or search when observed use justifies them.

  4. Later · Evaluate callable tools

    Explore a useful Workbench operation or MCP integration only after a repeatable user need emerges. No MCP service is part of this release.