If someone asked whether AI-assisted development has actually saved your team time, could you answer with confidence? Writing the change is only one part of the job. Every change still has to be reviewed, tested, deployed and supported once it’s live. If more of those changes are breaking, or coming back for rework, the time AI saves can quietly disappear further down the line.
The trouble is that most of the ways teams judge AI’s impact measure activity instead of outcomes. In this article, I’ll look at why the push for output is skewing how we judge AI’s value, what it’s doing to org stability and the people behind it, and how to find out whether AI is really making your team more effective.
The Obsession With Output
Organizations have become obsessed with output and the speed at which their teams can produce more. I’ve always been a fan of the “less is more” philosophy, especially when it comes to Salesforce orgs. But AI has fundamentally changed how we perceive value. Companies are increasingly hell-bent on trying to max out change output as a way of justifying AI platform usage and token spend.
This take isn’t just a gut-feel. Google’s DORA (DevOps Research and Assessment) team uncovered in their State of AI-Assisted Software Development Report 2025 that teams are feeling mounting pressure from leadership to increase output and deliver to tighter deadlines due to the perceived capacity gains from using AI.
I’m not saying that more output is bad, but the push for increased output tends to translate into tracking lines of code contributed, tickets closed, pull requests merged or features shipped – all of which can be quite easily gamed. So chasing output for output’s sake is meaningless.
If we’re actually going to see benefits from increasing output, we need to consider how effectively that output gets to production and how it performs when it’s there.
Rising Code Instability
Another widely held expectation is that admins and developers will have more time to focus on the most important work, now that AI can deliver the lower-level tasks. And AI can free up capacity, but that doesn’t happen automatically – in fact, it’s the exception and not the rule.
While the perception may be that AI helps us develop and fix things faster, DORA’s latest report actually found that increasing use of AI-assisted development is strongly correlated with a rise in org instability. Rising instability then increases the amount of time needed for rework, so the tool that has been sold to us on the grounds of increasing efficiency is cannibalizing more of our time.
You don’t need a crystal ball to see where I’m going with this. More output is fruitless if you’re sacrificing stability for throughput, and sinking the time you’ve saved into fixing the changes AI has generated.
(Re)Introducing the New and Improved DORA Metrics
This is why I think it’s time for us to reintroduce Google’s DORA metrics. These metrics were established as a way to understand how effective a team’s release process (more specifically, a DevOps release process) is by measuring both throughput and stability.
The metrics are:
- Deployment frequency: How often your team deploys to production.
- Lead time to change: The time taken between a ticket being started and deployed to production.
- Change failure rate: Percentage of deployments that fail, cause an incident, or need to be rolled back.
- Time to recover: Time taken to restore service following an incident or outage.
For over a decade, the DORA metrics have been the gold standard for teams looking to accurately measure how effective their DevOps processes are.
But with the rising need for rework and the increase in code instability mentioned earlier, the DORA team knew the metrics didn’t quite capture the full picture of throughput versus stability in an AI-driven world. So they introduced a new metric in 2024:
- Rework rate: ratio of unplanned deployments or changes requiring rework.
With rework rates now part of the DORA metrics, teams can accurately understand whether their use of AI is making them more effective or whether the push for rapid delivery is coming back to bite them further down the line. Ultimately, it stops teams from getting a skewed perception of the value that AI is adding for their team.
Keeping It Human: Beyond the Quantitative
I can’t not supplement this with the human side of things too. AI can add serious value, and pipeline data is great, but you can’t forget the people behind it all. Here are two key ways to make sure you’re still centering your colleagues in your decision-making:
Be Mindful of When Less Is More
When delivering changes becomes “cheap”, it’s tempting to do everything. And with AI-assisted development, it can seem easier than ever to deliver more changes and ask for work simply because it can be delivered more quickly rather than because it’ll add meaningful value. The unsurprising knock-on effect of this is that teams are reporting significantly increased levels of burnout.
An effective guardrail against falling into this trap is remembering that code is a liability, not an asset. Every new line of code introduces a new element of risk that needs to be managed long-term.
So it really does ring true that sometimes the cheapest and easiest change to manage is the one that was never made at all – and that applies to the cost to your business, your org, and your teammates.
Always Seek Out the Human Context
While monitoring the DORA metrics can show you what is happening, it can’t explain why. Low deployment frequency could be caused by tech debt, bureaucracy, or a burnt-out team. The numbers don’t tell you which.
So, while you may have a DORA “scorecard”, keep an eye on the human signals too:
- Trust in AI: 30% of people report little or no trust in AI-generated output.
- Verification load: process friction has shifted from building to verifying AI-generated output.
- User outcomes: DORA has found AI can hurt performance when teams aren’t focused on their people. Are your users happy, and are features being adopted?
AI is an amplifier for your Salesforce team, and we need to decide what to amplify rather than simply creating more white noise. Let’s measure whether our systems are delivering, and ask our teams inside the system how it’s feeling. This is the path to success.
Get AI-Generated Changes Ready for Review, Not Rework
If rework on AI-generated changes is cancelling out your time savings, the fastest fix is AI tooling that’s fit for purpose. Relying on review to catch poor AI output is risky and inefficient, and it’s a quick way to burn out your team. The changes entering your pipeline need to be right from the first iteration.
Cam, Gearset’s AI agent for Salesforce, is built to get the change right from the start. It has built-in practitioner skills, automatically stays up to date with your org’s metadata, and follows your team’s existing patterns.
When a ticket is unclear, it asks clarifying questions instead of building on assumptions. Your team also accesses the same agent, ensuring consistency in output quality rather than relying on how individuals have trained their own instances.
Every change is pre-validated against your org before your team is looped in for review, and if the change hits problems downstream, Cam will fix the issue. No maintenance or ramp time involved, and consistent testing and coding standards are applied for every change regardless of who made it.
You don’t have to take Cam’s value on trust. Gearset’s DevOps performance dashboard tracks your DORA metrics directly from your pipeline: deployment frequency, lead time for changes, time to restore, and change failure rate based on the bugs your team flags against deployments. Each period is compared with the one before it, so you can see whether throughput and stability are moving in the right direction as Cam takes on more of your backlog.
Curious to see Cam in action? Check out our on-demand webinar ‘From prompting to delegating: How Salesforce teams unlock real AI capacity.’
