Session Insights

Analyze your Devin sessions and get actionable feedback to improve future interactions

What is Session Insights?

Session Insights is an analysis feature that helps you understand what happened in your Devin sessions and provides actionable recommendations for improvement. When you trigger an analysis, Session Insights examines the session to identify patterns, issues, and opportunities for better collaboration.

Session Insights is available for all completed Devin sessions at no additional cost. Insights are not generated automatically — you must trigger analysis either through the UI or via the API.

How to Access Session Insights

Step 1: Complete a Session

Run a Devin session and let it complete. Session Insights works best with sessions that have clear outcomes, whether successful or not. Sessions that are too short (fewer than one Devin message) will not generate insights.

Step 2: Open the Insights Modal

After your session completes, look for the Session Insights button in the top bar of your session.

Step 3: Generate or View Analysis

Click the button to open the Session Insights modal. If an analysis has not yet been generated, click Generate Analysis to start one. Generation typically takes about a minute. If an analysis already exists, you can click Regenerate to create a fresh analysis.

Session Overview Metrics

At the top of the Session Insights modal, four key metrics give you a quick snapshot of the session:

ACU Usage

ACU (Agent Compute Unit) usage reflects how much compute Devin consumed during the session. Lower ACU usage for a given task generally indicates a more efficient session. Use this metric to compare similar tasks and identify sessions where Devin may have spent excessive compute on retries or dead ends.

User Messages

The total number of messages you sent during the session. A high message count can indicate that Devin needed frequent course corrections, suggesting that the initial prompt could be more detailed. Ideally, provide all important context upfront to minimize back-and-forth.

Session Size

Session size is a composite classification (XS, S, M, L, XL) based on both ACU usage and user message count. Either higher ACU usage or a higher number of user messages can increase the session size.

The thresholds for each size category are:

Size	ACU threshold	User message threshold
XS	≤ 2 ACUs	≤ 2 messages
S	≤ 5 ACUs	≤ 5 messages
M	≤ 10 ACUs	≤ 10 messages
L	≤ 20 ACUs	≤ 20 messages
XL	> 20 ACUs	> 20 messages

For enterprise customers, ACU thresholds are scaled by a factor of 10 (e.g., XS ≤ 20 ACUs, S ≤ 50, M ≤ 100, L ≤ 200, XL > 200). User message thresholds remain the same.

The overall session size is the larger of the ACU-based size and the message-based size. Sessions classified as L or XL are flagged as unhealthy, meaning Devin likely encountered significant issues or the task scope was too broad for a single session. Consider breaking large tasks into smaller, focused sessions.

To keep sessions small and efficient, provide all important information upfront in the initial prompt.

Category

Devin automatically classifies sessions into task categories based on the work performed. Classification also includes metadata such as the tools and frameworks used and the programming languages involved.

The available task categories are:

• Feature Development — building new functionality
• Bug Fixing — diagnosing and resolving bugs
• Code Review & Analysis — reviewing or analyzing existing code
• Refactoring & Optimization — improving code structure or performance
• Test Generation — creating unit tests or test suites
• Migrations & Upgrades — upgrading dependencies or migrating systems
• CI/CD & DevOps — working with pipelines, deployment, or infrastructure
• Code Quality & Security — addressing linting, security, or quality issues
• Data & Automation — data processing or automation scripts

This classification helps you understand how Devin interpreted your task and can reveal misalignment between what you intended and what Devin worked on.

Analysis Tabs

The Session Insights modal contains three tabs, each focused on a different aspect of the analysis.

Issue Timeline

The Issue Timeline tab contains two sections:

Issues Detected lists problems Devin encountered during the session. Each issue includes:

• A label describing the issue category
• An impact rating (high, medium, or low)
• A description explaining what went wrong

Issues are grouped by label and impact level, making it easy to see patterns. Common issue types include build failures, environment configuration problems, incorrect assumptions about the codebase, and scope ambiguity.

Timeline provides a chronological, color-coded view of key events during the session:

Color	Meaning
Red	High impact issue
Yellow	Medium impact issue
White/Gray	Significant event
Green	Value provided

Each timeline event has a title and description. Events linked to specific issues appear in bold. Use the timeline to understand the flow of the session — where Devin made progress, where it hit obstacles, and how it recovered.

Actionable Feedback

The Actionable Feedback tab helps you improve future sessions in two ways:

Improved Prompt shows a rewritten version of your original prompt with specific improvements. The suggested prompt is displayed with interactive highlighting — hover over an underlined section to see what changed and why. A numbered list of Changes Made below the prompt explains each modification:

• Added context or constraints that were missing from the original
• Clarified ambiguous instructions
• Included success criteria or specific requirements
• Frontloaded important information that Devin needed earlier

Click Start new session to launch a new Devin session pre-filled with the improved prompt.

Action Items lists recommended configuration changes to improve future sessions. These are concrete steps you can take in your environment configuration or Knowledge setup:

• Machine setup — environment or tooling changes (e.g., installing missing dependencies, configuring access)
• Repo config — repository-level changes (e.g., adding build scripts, updating configuration files)

Click Go to machine to navigate directly to your machine configuration and apply the suggested changes.

Knowledge Usage

The Knowledge Usage tab shows how your Knowledge items were used during the session:

Useful Knowledge lists knowledge items that helped Devin complete the task successfully, with an explanation of how each piece of knowledge was applied.

Misleading Knowledge lists knowledge items that led Devin astray or contained outdated or incorrect information. Each entry explains why the knowledge was harmful, helping you identify items that need updating or removal.

Click on any knowledge item to navigate directly to it and make edits. Regularly reviewing this tab helps you maintain a high-quality knowledge base.

Interpreting Common Insight Patterns

High ACU Usage with Few User Messages

This typically means Devin worked autonomously but struggled with the task. Check the Issue Timeline for recurring errors or retries. Common causes:

• Missing environment setup (dependencies, API keys, access credentials)
• Ambiguous requirements that led to trial-and-error approaches
• Complex tasks that would benefit from being broken into subtasks

What to do: Review the Improved Prompt for suggestions on adding context. Check Action Items for machine or repo configuration changes.

Many User Messages with Low ACU Usage

This suggests frequent interruptions or course corrections. Devin spent little compute but needed constant guidance. Common causes:

• Underspecified initial prompt
• Devin misunderstood the task scope or requirements
• The task required domain-specific knowledge not available to Devin

What to do: Use the Improved Prompt as a template for future similar tasks. Add relevant details to your Knowledge so Devin can access them automatically.

Misleading Knowledge Flagged

When the Knowledge Usage tab shows misleading knowledge items, those items may contain outdated instructions or overly broad advice that conflicts with your current codebase. Common causes:

• Knowledge was written for a previous version of your codebase
• Knowledge is too general and gets retrieved in irrelevant contexts
• Knowledge conflicts with other knowledge items

What to do: Update or delete the flagged knowledge items. Make knowledge trigger descriptions more specific to avoid irrelevant retrieval.

Session Classified as Wrong Category

If the category shown in the overview does not match what you intended, it likely means Devin interpreted your request differently. Common causes:

• The prompt was ambiguous about the goal
• The task description focused on one aspect but the intent was different (e.g., describing a bug when you wanted a feature)

What to do: Compare the category with your intent. Use the Improved Prompt to see how the analysis recommends clarifying the task objective.

Timeline Shows Repeated Issues

When the same issue type appears multiple times in the timeline, Devin likely got stuck in a retry loop. Common causes:

• A persistent build or test failure that Devin could not resolve
• An environment issue (missing tool, wrong version, permission error)
• A fundamental misunderstanding of the approach needed

What to do: Check Action Items for environment fixes. Consider adding a Knowledge item that explains the correct approach for this type of task.

Best Practices

Review Insights After Complex Sessions

Make it a habit to check Session Insights after important or complex sessions. The patterns you identify will help you become more effective over time.

Apply Prompt Improvements Iteratively

Use the suggested improved prompts as starting points for similar future tasks. Over time, you will develop a library of effective prompt patterns. Save your best prompts as Playbooks for repeatable workflows.

Maintain Your Knowledge Base

Regularly review the Knowledge Usage tab to keep your knowledge items accurate and relevant. Remove or update misleading knowledge promptly — a single outdated knowledge item can degrade session quality across your entire team.

Address Recurring Issues via Machine Setup

If Action Items consistently recommend the same environment or configuration changes, address them proactively. Setting up your environment configuration correctly prevents repeated issues across all future sessions.

Share Insights with Your Team

Session Insights can reveal patterns that benefit your entire organization. Add key learnings as Knowledge so your teammates can benefit from them.

Keep Sessions Focused

If your sessions consistently classify as L or XL, break large tasks into smaller, more focused sessions. Smaller sessions tend to produce better results and are easier to analyze and iterate on.

Troubleshooting

No Insights Available

If Session Insights is not available for a session, it may be because:

• Analysis has not been triggered yet — click Generate Analysis in the Session Insights modal or use the generate API endpoint
• The session is still in progress
• The session was too short to generate meaningful analysis (fewer than one Devin message)
• There was an error during the analysis process — try clicking Regenerate

Analysis Takes Too Long

Analysis generation typically completes within a minute. If it has been generating for more than five minutes, the process may have timed out. Close and reopen the Session Insights modal, then click Regenerate.

Investigate with Devin

On Team and Enterprise plans, users with the appropriate permissions will see an Investigate with Devin button in the Session Insights modal. Clicking this button opens a new Devin session pre-configured to analyze the original session in depth. Use this for sessions where the automated analysis alone does not fully explain what happened.