A Case Study: Tesla Stock Investment Report

An Identical Prompt Used to Compare the Implementation Results of Manus and Devin

(Content from Kenny, Product Lead at Impa)

The general-purpose AI Agent that launched earlier this morning has quickly become a trending topic within the AI community. Within hours, invitation codes became highly coveted, driving a surge of users to Discord in hopes of securing one from Tao Cheung.

Manus's official website features 40 case studies demonstrating its capabilities across a broad range of scenarios, including travel planning, stock analysis, curriculum design, insurance policy comparisons, B2B supplier sourcing, financial report analysis, company data compilation, online store operations analysis, diagram generation, candidate interview scheduling, lead generation, and teleprompter scriptwriting for press conferences.

From the use cases presented, it’s evident that Manus’s interface design bears a strong resemblance to last year’s breakout success, Devin. Both feature modular components such as Browser, Shell, Editor, and Planner.

Devin: A Brief Overview

Devin, developed by Cognition AI, is positioned as an autonomous software engineer capable of completing full-cycle development tasks—from coding to deployment—with minimal user intervention. Users simply provide a prompt and receive periodic updates, unlike many other tools that require constant supervision and interaction.

Currently, Devin is priced at $500/month for team access and is not available to individual users. In contrast, Manus remains in closed beta, with public testing yet to open. Nonetheless, early demo footage and official case showcases are already accessible online.

01 Case Study: Investment Analysis of Tesla Stock

As a newly introduced AI Agent, Manus shares many functional similarities with Devin. To evaluate their performance and compare the user experience, we conducted a side-by-side test using an investment analysis task focused on Tesla stock—applying the exact same prompt for both agents.

In this case, the AI Agent was expected to showcase its ability to autonomously retrieve Tesla’s historical stock data, conduct technical and financial analysis, incorporate market trends and news sentiment, and ultimately provide detailed investment recommendations. Additionally, the agent needed to demonstrate the ability to package and publish the final analysis as a publicly accessible website.

We used the initial prompt exactly as shown on Manus’s official site, with no modifications made to the conversation or publishing process:

The user interface of Manus’s official platform is shown below:

Devin's interface used the same prompt:

Throughout the process, both Manus and Devin generated a task plan via a TODO list. Devin embedded its plan directly into the user interface, while Manus generated a markdown file to track the task status. In daily usage by the Impa team, Devin also frequently creates todo.md files to monitor the progress of specific execution plans.

Ultimately, both Manus and Devin delivered comprehensive reports with charts and detailed content:

After the analysis was completed and deployed to a public-facing site, this was the final web output for Manus:

And here is the final output from Devin:

It’s evident that Manus delivers a more polished and professional report presentation. Its structure features an Executive Summary and Recommendations at the top, following a clear top-down logic. Visual charts and data modules are grouped at the bottom. By contrast, Devin’s layout seems less polished in color palette and chart design, with its overall narrative structure and data visualization showing less technical finesse.

That said, both AI Agents include advanced features such as Knowledge Base and Memory, so it’s possible that these case demonstrations were enhanced by additional internal knowledge support.

02 Prompt Optimization Based on Agent Capabilities

One notable advantage of Devin is its built-in prompt optimization. Upon receiving an initial instruction, Devin can automatically refine the prompt—prompting the user to clarify data sources, APIs, tools, and even suggesting task decomposition for better execution. ChatGPT’s Deep Research feature uses a similar confirmation mechanism, ensuring greater precision by validating key details before execution begins.

Below is the optimized prompt version generated by Devin:

The final website generated using this optimized prompt features pagination and interactive charts, resembling a frontend web application:

From a technical stack perspective, the websites built by Manus and Devin using the original prompt were nearly identical. Neither utilized a modern front-end framework, and both relied on Chart.js, a lightweight charting library.

From a technical stack perspective, the websites built by Manus and Devin using the original prompt were nearly identical. Neither utilized a modern front-end framework, and both relied on Chart.js, a lightweight charting library.

This case study illustrates just one example among many use cases for AI Agents. Despite their shared foundation in large language models, Manus and Devin each reflect distinct design philosophies and are suited for different types of workflows.

Manus: A Product Design Driven by Smart Analysis

Manus acts as a highly efficient assistant focused on actionable task completion and practical results. When given a concrete directive, it excels at delivering intuitive, data-rich analysis reports—ideal for users who need immediate insights.

Devin: A Product Designed with an Engineering Mindset

Devin, in contrast, is built as a “junior developer” for software teams. Beyond task execution, it supports engineering workflows such as Slack command integration, GitHub PR automation, secret management, in-editor code control, and seamless compatibility with tools like Cursor. These features make Devin particularly well-suited for long-term system development and maintenance.

As large language models continue to evolve, AI Agent functionality is being increasingly integrated into the core capabilities of LLMs themselves. Future products will likely strike a better balance between quick results and engineering depth. Manus and Devin each offer unique advantages, and teams should choose based on their specific use case. In time, next-generation models may consolidate both strengths into a unified, intelligent experience.

As a crypto-native investment fund, Impa Ventures closely monitors the productivity gains and business transformation potential of LLMs—particularly in the Crypto/Web3 ecosystem. For more on what the “code surplus” era means for us, read: What Does Code Abundance Mean for the World?.