Which AI coding assistant offers the highest ROI? Here's how to find out

AI transformation leaders rely on Faros to navigate critical decisions in AI adoption, impact, and ROI. And as the AI coding landscape evolves, models improve, and the tools become more powerful, new questions emerge. 

We're announcing strategic additions to our industry-leading AI Transformation product to help answer your most critical questions. These include advanced metrics to measure productivity gains from AI coding tools that every engineering leader needs: 

• Which developer usage patterns actually drive productivity improvements? 
• How much of our codebase is being written by agents? 
• Which AI coding tools, features, and foundational models are worth paying for?

{{cta}}

This product release honors Toru Iwatani, creator of Pac-Man, whose pioneering ghost algorithms established distinct AI personalities in gaming. The four ghosts collaborate without explicit coordination, with their individual patterns naturally forming team tactics. Though not "intelligent" by today's standards, they responded dynamically to player behavior, creating an illusion of personality and adaptability that mirrors modern human-centered AI principles.

The essence of Iwatani's design—AI that feels alive, collaborative, and responsive—mirrors how modern AI systems aim to work with people, not just for them.

Let's dive in.

Metrics to measure productivity gains from AI coding tools: Which usage patterns drive impact?

You've rolled out AI coding tools across your organization: GitHub Copilot, Cursor, Claude Code, Windsurf, Augment, Devin, and others. But here's the real question: Which developer usage behaviors actually move the needle on productivity metrics, like velocity and quality?

The Iwatani Release introduces rich developer behavior insights that connect tool usage patterns to engineering outcomes. 

Which AI coding tool is most popular with my developers? 

Start by answering this question: From all the AI coding assistants at their disposal, which tool do your developers prefer? 

Faros measures adoption across all the AI coding tools in your stack, so you can instantly see where their preference lies. Usage data is available at every level of your organization, all the way down to the individual team.  

Key questions AI transformation leaders should ask when analyzing AI coding assistant popularity:

• What drives developer preference for specific AI tools over others?
• How can we align tool selection with the unique needs of different product areas and team structures?
• Are our current enablement investments sufficient to maximize adoption and value?

Which AI coding tool features provide the most value?

Understanding tool preference is just the first step. There’s more to learn from digging deeper into the specific features developers actually use within each tool. This data reveals what developers find most valuable and where they see the biggest benefits.

For each of your AI coding tools, Faros measures the usage of its capabilities, which may include: 

• Autocomplete: inline code suggestions as you type
• Code generation: create code from natural language prompts
• Chat / Q&A: conversational help and explanations
• Code analysis & review: error detection, refactoring, optimization
• Context awareness: help with project-level or multi-file-level understanding
• Agentic: autonomously handle a PR or task 
• Documentation & explanation: describe or generate docs for code
• Dev workflow integration: tool actions, commits and PRs

Key questions AI transformation leaders should ask when analyzing AI coding tool feature-level usage:

• Which tool is preferred for each function, for example, which tool is most popular for code reviews?
• Which tool’s agentic mode is most trusted?
• Which tool is best for providing project context across multiple repos? 
• What's driving low adoption for certain features? Try to determine if the root cause is inadequate training, restricted access permissions, or simply that the feature doesn't provide enough value to users.

How does AI coding tool usage frequency impact productivity? 

The main goal of AI tools is to boost engineering productivity, so it's important to figure out how often developers need to use them to see real benefits. In other words, what's the minimum usage frequency required for developers to experience clear improvements in their speed, output, and code quality?

First, Faros allows you to see at a glance how AI coding tool usage is progressing over time. The data is clearly visualized across a timeline, revealing the trends and inflection points driving AI adoption.

Next, Faros connects usage frequency with productivity and quality metrics to find the sweet spot where AI adoption creates real, measurable improvements. You can compare the impact across five categories: no usage, infrequent usage, moderate usage, frequent usage, and power usage. 

In the chart examples featured below, you can see how different usage levels impact key velocity metrics. 

Key questions AI transformation leaders should ask when analyzing the correlation between adoption and impact:

• What is the minimal usage frequency to target with your developers? For example, you may discover that moderate usage and frequent usage result in the same impact gains. Alternatively, perhaps only frequent users see improvement. 
• Where are the pockets of non-usage and infrequent usage in your org? Meet with these teams to understand their blockers. 
• How much more productive are power users vs. moderate users? Interview your power users to understand how to increase usage among other developers. 

{{cta}}

AI agents are writing more code: What is the impact?

Engineering leaders are increasingly seeking to understand how the nature of software development is shifting as AI agents begin to play an active role in their codebases. 

With major organizations like Meta and Microsoft reporting that roughly 30% of their code is now AI-generated, the question is no longer if AI is reshaping software engineering, but how much of it AI now drives.

The Iwatani Release gives engineering leaders the visibility they need to answer three key questions:

• What portion of our codebase has been written by AI?
• How much productivity do AI agents contribute to my org?
• How much rework is AI introducing?

What portion of our codebase has been written by AI?

Is this code AI-generated? Everyone wants to know. Understanding AI’s footprint is critical, because it directly affects:

• Long-term codebase viability (maintainability, security, and compliance)
• Code quality (readability, organization, and ongoing monitoring)
• Strategic workforce implications (mentorship, training, and the propagation of best practices)
• Developer evolution (how engineers grow and adapt in an AI-driven environment)

Now, AI transformation leaders can measure the percentage of your codebase authored by AI, with detailed views per AI tool, repository, and team.

By making AI’s contribution measurable, leaders can manage risk, maintain quality, and plan for how teams will evolve alongside their automated counterparts.

Here's some guidance for AI transformation leaders as they examine this data:

• Where are we accumulating hidden risks, and how do we proactively address them? Assess whether your code review processes are adequately equipped to handle repositories where 25%+ of code is AI-generated, and whether additional scrutiny protocols should be implemented.
• How is AI usage affecting our developers' skill development and team capabilities? Examine whether developers maintain their ability to read, understand, and troubleshoot AI-generated code, and whether the organization needs to implement training or mentorship programs to preserve critical coding skills alongside AI adoption.
• How do we optimize our AI governance and quality assurance processes based on these usage patterns? Decide which repositories or code types require enhanced review processes, how to balance AI efficiency with code quality standards, and whether current testing and security practices are sufficient for the actual level of AI integration across different parts of the codebase.

How much productivity do AI agents contribute to my org?

Now you can quantify how much productivity AI agents are adding to your organization, with drill-downs available by AI tool, repository, and team. The Iwatani Release helps you understand:

• How many PRs are authored by AI agents
• How many PRs are reviewed by AI agents 

These insights let you compare the units of productivity AI is contributing to inform future capacity planning. Key questions AI transformation leaders should ask when measuring AI agent contributions:

• How many additional AI agents do we need to hit our roadmap targets?
• Could more AI-powered code reviews reduce our cycle time bottlenecks?
• What should our future engineering team structure look like with AI agents?

How much rework is AI introducing?

Not all AI-generated code leads to lasting gains. Some of it introduces rework, a reflection of both the quality of AI contributions and their human ramifications. Faros enables orgs to track rework rate (the fifth DORA metric) to see where AI-generated or AI-accelerated code is creating inefficiencies. 

These insights allow organizations to balance speed with quality, ensuring that AI-driven development results in real productivity improvements rather than hidden waste.

{{cta}}

ROI and investment insights: Which AI coding tools are worth paying for?

Once you understand AI coding assistant usage and can measure the impact, you have all the essential metrics to measure productivity gains from AI coding tools and decide which investments are worth paying for. Most importantly, you can finally answer the critical question: which AI coding assistant offers the highest ROI for your organization?

When every model, feature, and token tier carries a cost, knowing which tools truly deliver ROI becomes essential. With Faros's Iwatani Release, the calculation has gotten even more sophisticated. 

These are the insights that help AI transformation leaders prioritize renewals and upgrades for tools that drive measurable outcomes, identify underused or low-impact features to optimize license spend, and inform vendor negotiations with data on what’s actually working.

How can I make informed model choices for our AI coding tools?

New models emerge, old ones get deprecated, performance and token consumption fluctuates. Having insight into the models your developers prefer, how much they cost, and what the tradeoffs are for cheaper models is extremely helpful. 

As a first step, Faros measures which models are used most often per tool.

Then we dig deeper; For any given tool, Faros shows you which model developers choose for each specific feature. These insights help you identify which models work best for different tasks and recommend the most cost-effective, high-performing options to your teams.

How much are we paying for models and how can we be more cost-efficient? 

Most AI coding tools operate on a token-based pricing model, where every interaction consumes tokens. The more tokens used, the higher the cost. 

In the Iwatani release, Faros introduced token consumption and cost tracking. Claude Code is the first AI coding tool to expose token usage and cost data through its API. Faros ingests this data automatically, giving you unprecedented visibility into the true cost of AI-generated code. See cost per commit, cost per feature, and total spend by team and repository. Evaluate when it’s worth moving up a tier for greater token capacity at a lower effective rate.

You can also get value-per-dollar calculations to determine which AI coding assistant offers the highest ROI. Which tools provide the best bang for your buck? Faros calculates cost efficiency ratios, like tokens consumed per commit or cost per productivity gain, so leaders can objectively compare tools and make budget decisions based on data, not vendor promises.

This is the financial clarity AI transformation leaders need: clear ROI metrics that justify AI investments to CFOs and help you optimize your tool portfolio for maximum impact at minimum cost.

Note: Faros easily connects to and ingests data from all major AI coding tools. For tools like Claude Code, GitHub Copilot, Cursor and Windsurf, which have exposed APIs, you can connect directly with a simple token. But for other AI coding tools, for example Augment and others, Faros can still capture the necessary data through alternative integrations, ensuring you can use those insights to build data-driven strategies regardless of which tools your teams use.

Beyond metrics: Diagnosing the adoption-impact gap

Having comprehensive AI adoption metrics is powerful, but what happens when high usage doesn't translate to productivity gains? When developers are actively using AI tools but cycle times haven't improved?

This adoption-impact gap is where many organizations get stuck. The issue usually isn't the tools—it's the organizational systems around them. Successful AI transformation requires strategic alignment across processes, culture, and measurement frameworks that goes beyond tool deployment.

GAINS™ bridges this gap. While the Iwatani Release delivers the metrics to measure productivity gains from AI coding tools, GAINS provides the strategic framework to act on those insights. Schedule your GAINS™ consultation today.  

Iwatani Release: AI transformation intelligence for every engineering leader

The latest release from Faros transforms how organizations measure, optimize, and invest in AI coding tools. From understanding developer behaviors to tracking agentic code contributions to making cost-informed investment decisions, Faros gives you the intelligence you need to lead confidently through the AI transformation.

AI is everywhere. With Faros, impact is too.

Ready to see how Faros can transform your AI strategy? Book your demo today.