Agentic Research Workflows

One model answers questions; a system of models changes how you think.

From prompting to orchestration

Most research conversations with AI follow a simple pattern: you write a prompt, get a response, iterate. This works well for individual tasks but breaks down for complex, multi-stage research projects.

Agentic workflows change the model. Instead of one researcher talking to one AI, you design a system of specialized agents that collaborate under your direction, each handling what it does best while you maintain strategic control.

If you want to locate your own practice before reading about the stack, Composing an Agentic Research System gives seven distinctions and five questions that do not depend on any particular tool.

        TRADITIONAL

+------------+     +-----------------+
| Researcher | <-> | Single AI model |
+------------+     +-----------------+

          AGENTIC

+------------+     +---------------------+
| Researcher | --> | Orchestration layer |
+------------+     +----------+----------+
                              |
             +---------+------+------+
             |         |      |      |
             v         v      v      v
         analysis  challenge  synthesis  memory

The agentic research stack

These components illustrate different ways to arrange agentic research, not one stack to install in full. The Toolkit index distinguishes tools used in routine practice from onboarding, experiments, and external-facing projects.

How the stack works together

The original integrated example remains useful as a map of how the tools relate. The AI Model Reference carries the current model choices.

Phase 1: foundation building

• Interpretive Orchestration enforces solo coding of initial documents
• Memex captures your emerging framework and theoretical sensitivity
• Vox lets you consult multiple models about methodological choices

Phase 2: collaborative analysis

• Interpretive Orchestration activates @dialogical-coder for parallel analysis
• Vox enables multi-model triangulation on findings that need cross-checking
• Kimi K3 (via Vox) stress-tests your emerging theory
• GLM-5.2 (via Vox or direct provider config) handles long-context coding-agent and multilingual synthesis checks
• Memex preserves the evolving analysis across sessions

Phase 3: synthesis and writing

• Memex recall / curator skills and memex ask surface patterns across all sessions
• Interpretive Orchestration @scholarly-companion examines work through an epistemological lens
• Claude Opus 5 (via Vox) helps break frames and find insights beyond local optima

Current practice: a shared project across hosts

The overall workflow remains repository-centered. Kimi is a regular collaborator alongside the models the author works with through Codex and Claude Code. The work spans reasoning, writing, coding, and sustained projects. What matters most is the difference in perspective: Kimi often notices something else or takes another approach.

The host is the interface; the model is a participant in the work. Working with Kimi through Codex or a plugin still brings Kimi's perspective into the project. Native Kimi Code is another route. The diagram shows the available arrangement; each task can involve a different combination.

+------------+     choose / approve / delegate
| Researcher |--------------------------------------+
+------------+                                      |
                                                    v
            +-------------+  +-------------+  +-------------+
            | Codex       |  | Claude Code |  | Kimi Code   |
            | + subagents |  | + subagents |  | + subagents |
            +------+------+  +------+------+  +------+------+
                   \                |                /
                    +---------------+---------------+
                                    |
                         +-----------v---------------+
                         | Shared project repository |
                         | plans / code / reports    |
                         +-----------+---------------+
                                    ^
                                    |
+------------------+    agent files report
| Traditional chat |----------------+
+------------------+

Traditional chat remains useful for exploration, discussion, and bounded tasks. In the author's current practice, multi-agent orchestration is the default for complex work. Depending on the project, the researcher may design the workflow directly, revise an agent's proposal, or delegate workflow construction to a lead agent.

For multi-model tasks that need agents to inspect files and carry out work, the author now uses cc-fleet (opens in a new tab) more often. Vox remains useful for one-shot consultations and defined prompt workflows. Both Codex Plugin for Claude Code (opens in a new tab) and Kimi Plugin remain part of the working setup. Lotus Wisdom is used more often than Sequential Thinking.

The project-opening invitation records a related prompting practice: invite agents to question the framing and surface possibilities beyond it, then decide together whether to follow them.

The durable record is distributed rather than confined to one application:

                      +----------------------------+
                      | Distributed research state |
                      +--------------+-------------+
                                     |
          +--------------------------+--------------------------+
          |                |                    |               |
          v                v                    v               v
+---------+------+ +-------+--------+  +--------+------+ +------+-------------+
| Zotero         | | Markdown files |  | Memex         | | Project repository |
| sources + refs | |       ^        |  | session memory| | plans/code/reports |
+----------------+ |       |        |  +---------------+ +--------------------+
                   | Obsidian       |
                   | interface      |
                   +----------------+

Zotero carries sources and bibliographic records. Obsidian provides an interface for Markdown files. Memex preserves cross-session reasoning and evolving interpretation. The project repository carries the operational artifacts that agents share.

Let Kimi frame the problem

For example, begin a methods or implementation task with Kimi. Share the question, source material, constraints, and criteria. Ask Kimi to propose an approach and identify assumptions worth questioning.

Then invite Claude or a GPT model to develop a separate account from the same brief before reading Kimi's proposal. Bring the accounts together and examine what each noticed, missed, or made possible. Decide which direction to follow, or develop another one together. Kimi can carry the work forward as well as critique it.

This is a suggested working pattern. The roles can reverse on the next task. Different perspectives are a reason to work across models; agreement alone does not establish that an account is sound.

Broader agentic research tools

Researchers can combine the ecosystem above with adjacent agentic tools:

CLI-based agents

• Codex, Claude Code, and Kimi Code: workspaces connected through shared project files. Kimi models participate both directly and through other hosts; see ways to work with Kimi.
• Antigravity CLI: Google's current terminal-agent path for former consumer/free Gemini CLI workflows
• OpenCode: Open-source, model-agnostic terminal agent

Open literature index

• OpenAlex: the API and CLI an agent searches the literature through, and the layer beneath Research Scanner

MCP servers

The Model Context Protocol (opens in a new tab) has grown to 1,000+ community servers, now managed by the Linux Foundation. Research-relevant servers include:

• Sequential Thinking: Step-by-step reasoning for any model
• Lotus Wisdom: Contemplative problem-solving
• MinerU: PDF/document parsing for large batches

Research platforms

These are external-facing projects rather than required stages in the author's daily workflow:

• Research Scanner: Literature surveillance pipeline balancing exploitation and exploration
• OpenInterviewer: AI-assisted qualitative interviews for larger studies

Getting started with agentic research

You do not need the full stack to benefit from agentic workflows. Each component works independently, so start with the part addressing your largest pain point and expand from there. The advanced path adds coordination without invalidating the original route.