# Session 4: Advanced Agentic Workflows

URL: https://research-memex.org/docs/case-studies/systematic-reviews/session-4-agentic-workflows
Description: Agentic workflows work only when each AI role has a task, a boundary, and a verification gate.



*Agentic design is the architecture of thinking at scale.*

We now turn to designing <Tooltip tip="Systems where AI agents can act autonomously within defined boundaries and quality controls">agentic systems</Tooltip> for complex, multi-step research processes. This is about architectural thinking: how to give each agent a role, define quality gates, and keep the human researcher in control.

This session also develops the critical thinking to decide when such systems are appropriate, how to maintain quality control, and where human judgment remains irreplaceable. By the end of this guide, you'll move from one-off prompting to deliberate orchestration.

## Learning outcomes [#learning-outcomes]

By the end of this session, you'll be able to:

* Design [multi-agent workflows](/docs/implementation/agentic-ai-tools/claude-code-setup-guide) where different AI roles collaborate on complex research tasks
* Implement quality control frameworks that maintain research rigor in automated systems
* Critically evaluate when agentic approaches add value and when simpler methods work better
* Navigate the ethical and professional considerations of increasingly autonomous research systems
* Build practical, cost-effective workflows that integrate into your academic research practice

## Multi-agent research architecture [#multi-agent-research-architecture]

```text title="diagram:multi-agent-architecture"
           +----------------------+
           | Human architect      |
           | goal / roles / gates |
           +----------+-----------+
                      |
                delegates through
                  checkpoints
                      |
                      v
           specialized AI roles
                      |
+-----------+  +----------+  +-----------+  +----------+
| discovery |  | analysis |  | synthesis |  | critique |
+-----+-----+  +----+-----+  +-----+-----+  +----+-----+
      |             |              |             |
  relevance       logic         novelty      integration
      |             |              |             |
      +-------------+------+-------+-------------+
                             |
                       all gates pass
                             |
                             v
                  +----------+----------+
                  | validated synthesis |
                  +----------+----------+
                             |
                     feedback / refine
                             |
                             +--------> Human architect
```

<Info>
  Critical distinction: agentic workflows are not about removing humans from research. They are about strategic delegation of cognitive labor. The human architect maintains oversight, designs the system, and validates outputs. Quality control gates protect rigor when tasks become more automated.
</Info>

## Key concepts: the workshop prep kit [#key-concepts-the-workshop-prep-kit]

The readings below trace the arc from vision to practice: a high-level case for AI co-scientists, the engineering realities of multi-agent systems, practical implementation patterns, and how agentic workflows may reshape academic research.

* Conceptual architecture ("[Towards an AI co-scientist](https://arxiv.org/abs/2502.18864)"): the high-level vision, moving from single prompts to orchestrating a team of [specialized AI roles](/docs/case-studies/systematic-reviews/claude-code-slr-workflow#section-2-the-three-tier-agentic-workflow).
* Engineering reality ("[How we built our multi-agent research system](https://www.anthropic.com/engineering/multi-agent-research-system)"): the "what can go wrong" guide. Learning from professional teams helps anticipate and solve common problems.
* Practical implementation ("[Claude Code: Best practices](https://www.anthropic.com/engineering/claude-code-best-practices)"): the primary lab manual, with specific instructions and code patterns for hands-on work.
* Professional context ("[From Scarcity to Abundance](https://doi.org/10.5465/amj.2023.4006)"): how agentic workflows may change academic research practice.

## Building your research architecture: a step-by-step guide [#building-your-research-architecture-a-step-by-step-guide]

<Steps>
  <Step title="Part A: failure analysis" icon="triangle-exclamation">
    Review failure patterns first: before building a complex system, analyze the common failure modes documented in the [Failure Museum](/docs/implementation/core-references/failure-museum). Understanding what goes wrong is the foundation for building systems you can inspect.

    Identify quality-control checkpoints: these documented failures become checkpoints for the new system.
  </Step>

  <Step title="Part B: quality & ethics" icon="shield-check">
    Address ethical considerations: review institutional policies on AI use, data privacy, and proper documentation of AI assistance.

    Build a quality-control framework: define the roles for human and AI at each stage of the research process, such as discovery, curation, analysis, and writing. Specify the verification method for each stage.

    <Warning>
      Do not skip this. Ethical frameworks protect both you and your research integrity.
    </Warning>
  </Step>

  <Step title="Part C: multi-MCP orchestration" icon="network-wired">
    Build a research agent system: use a tool like Claude Code to connect multiple MCP servers, such as filesystem and web search.

    Chain servers for workflows: design workflows that pass information between different servers to solve problems identified in your failure analysis. This is where theory meets practice.
  </Step>
</Steps>

For a step-by-step implementation of an SLR with Claude Code, including screening, extraction, and synthesis, see [Building an SLR with Claude Code](/docs/case-studies/systematic-reviews/claude-code-slr-workflow).

***

<Card title="Recommended Exercise: Design Your Own Agentic Workflow" icon="lightbulb">
  1. Define your use case: build upon a systematic review or another research task.
  2. Design your architecture: map out which tasks will be automated vs. human-controlled.
  3. Specify your agents: define the roles and prompts for each AI agent in your system.
  4. Plan quality controls: identify verification points and failure modes, using your documented failures as a guide.
  5. Calculate costs: estimate API usage and time savings.

  This is the capstone exercise for applying agentic thinking to your own research.
</Card>

## Pre-class setup for Session 4 [#pre-class-setup-for-session-4]

Before this session, please ensure your research environment is fully prepared:

<Steps>
  <Step title="Curate your knowledge base">
    Have all your curated papers loaded and processed in your Cherry Studio knowledge base.
  </Step>

  <Step title="Test your workflow">
    Verify that the conversation export functionality to Obsidian is working correctly.
  </Step>

  <Step title="Prepare your protocol">
    Bring your draft systematic review protocol. We will be using it as the foundation for building our agentic workflows.
  </Step>
</Steps>

## Beyond the guide [#beyond-the-guide]

The workflows you design here can become the foundation for dissertation research, collaborative projects, and professional research practice.

The takeaway: agentic AI does not replace human expertise. It extends what the researcher can inspect, delegate, and verify.

## Go deeper [#go-deeper]

Ready to explore advanced AI capabilities?

<IndexRows>
  <IndexRow title="Agentic Workflows" href="/docs/advanced-topics/agentic-workflows">
    Build multi-agent research systems
  </IndexRow>

  <IndexRow title="MCP Servers" href="/docs/toolkit/sequential-thinking-mcp">
    Extend AI capabilities with tools
  </IndexRow>
</IndexRows>

## Related Resources [#related-resources]

<IndexRows>
  <IndexRow title="Setup Guides" href="/docs/implementation/foundational-setup/quick-start-checklist">
    Complete tool installation and configuration
  </IndexRow>

  <IndexRow title="Core References" href="/docs/implementation/core-references/cognitive-blueprint-prompts">
    Prompt templates and model guides
  </IndexRow>

  <IndexRow title="Failure Museum" href="/docs/implementation/core-references/failure-museum">
    Learn from common AI limitations
  </IndexRow>

  <IndexRow title="AI-Readable Docs" href="/docs/introduction/llms">
    llms.txt and Markdown access for AI tools
  </IndexRow>
</IndexRows>

***

Navigation: [Previous: Session 3](/docs/case-studies/systematic-reviews/session-3-human-vs-ai-synthesis) • [Return to Case Study Overview](/docs/case-studies/systematic-reviews/systematic-review-syllabus)