---
title: "Deep Research: Practitioner Research Tools"
created: 2026-09-03
visibility: public
status: live
type: research-summary
tags: [deep-research, ai-tools, workflows, claude-code]
summary: "What people who research for a living actually run, in January 2026. Nobody uses one tool, and the thing holding the stack together is context management."
---

I looked at what people who research for a living actually run.

Nobody uses one tool. Gathering, deep research, real-time social, and synthesis are four different jobs, and the people doing this well use four different tools for them. Perplexity for a fast cited answer in about five minutes. Gemini for structured analysis in twelve. ChatGPT when the question is exploratory. Grok when it has to be live from X, because it is the only one with that access. Claude Code for synthesis and anything touching code.

Boris Cherny runs five parallel Claude Code sessions locally and another five to ten on the web. His line is the keeper. Give Claude a way to verify its work and it doubles or triples the quality. He also throws away 10 to 20 percent of sessions that go nowhere, which is a relief to hear from the person who built the thing.

Addy Osmani works the other way around. One function, one bug, one feature at a time, with automated tools catching what the model misses and the model fixing what the tools catch.

What holds any of it together is not the model. It is context management. A CLAUDE.md in git, a directory for reasoning saved between phases, reference files an agent can reread on the next run. That is [[Business management is fundamentally the management of expanding context]] showing up as a tool stack.

Full Deep Research:
[[Practitioner Research Tools: What Smart Developers Actually Use (January 2026)]]