Most development teams building on Salesforce are using AI, but there’s stark variation between teams in terms of how much AI has impacted productivity, speed of delivery, and code quality. Many teams still doubt the accuracy and security of AI-generated changes. So what’s driving the difference in outcomes from adopting AI?
The State of Salesforce DevOps 2026 found that the strongest predictor for confident and effective use of AI was DevOps. In this article, we’ll see why DevOps underpins trusted AI, and highlight the areas to prioritize so you can build and ship quicker using AI without losing control.
AI-assisted development is widespread, but trust lags behind
At the time of the survey, 84% of teams were using AI to build Salesforce changes — and adoption has no doubt increased during the year. Teams use a range of tools: Agentforce Vibes, GitHub Copilot, ChatGPT, Gemini, and Claude.
While the tools of choice change, with Claude overtaking ChatGPT during the year among Salesforce teams, adoption is still outpacing trust. Only 40% of respondents are confident in the accuracy and security of AI-generated code and configuration.
Trust in AI varies depending on the task at hand. We asked teams how much they trusted AI for different stages of the DevOps lifecycle: plan, build, validate, release, operate, and observe. Build is where teams are most comfortable to hand over work to AI, with 82% happy for AI to assist or work autonomously. Release is where they’re most cautious: 42% want no AI involvement there yet, and operate isn’t far behind at 39%.
| How much teams trust AI at each stage | AI-assisted | Fully autonomous agents | No AI |
|---|---|---|---|
| Plan | 57% | 16% | 27% |
| Build | 71% | 11% | 18% |
| Validate | 60% | 12% | 28% |
| Release | 48% | 10% | 42% |
| Operate | 54% | 7% | 39% |
| Observe | 53% | 15% | 32% |
Full lifecycle coverage goes hand in hand with AI confidence
The DevOps lifecycle is the path every change takes, from planning the work to watching how it behaves in production. Most teams (73%) have every stage covered — either with tooling, defined process or both. The other 27% have a gap at one or more stages.
Those two groups use AI differently. Teams with full coverage are more likely to use AI tools for development in the first place (80% against 56%). And among teams that do use AI for development, the full-coverage group comes out ahead on every measure:
| Teams using AI for development | Every stage covered | Gaps in one or more stages |
|---|---|---|
| Confident in the security of AI-generated changes | 42% | 15% |
| Confident in the accuracy of AI-generated changes | 48% | 28% |
| Confident in the performance of AI-generated changes | 56% | 37% |
| Bugs in fewer than 5% of releases | 42% | 25% |
| Release to production daily or more often | 14% | 8% |
| Fix a typical production bug in under 2 hours | 31% | 24% |
| Restore service within 24 hours of an incident | 96% | 90% |
The security gap is the widest. Teams with every stage covered are nearly three times as likely to be confident in the security of AI-generated changes. These teams have reliable, scalable ways to check what AI produces, with automated tests and static code analysis run on every change regardless of who wrote it. And when a bad change slips through, monitoring in the operate and observe stages flags the error before a user reports it.
Discover the key insights for DevOps performance
The State of Salesforce DevOps
Close these gaps to build trust in AI
Testing and validation do the most to stop bugs and security vulnerabilities from reaching production, so should be non-negotiable for every team. At the same time, operate and observe are the stages teams most often have process and tooling gaps — and these are the stages most tied to confidence in AI.
The areas to prioritize are:
- Make every change traceable. With version control as the source of truth, an AI-generated change has an author, a history, and a way back. Salesforce has no native rollback, so knowing how to reverse a deployment matters more as change volume grows. Among teams using AI, those with build and release tooling are far more likely to have bugs in fewer than 5% of releases (41–42% against 23–25%).
- Automate validation. AI raises the volume of changes, and manual review doesn’t scale with it. As Capgemini’s Christopher Ramm writes in the report’s foreword, AI compresses the build stage, which turns validation into the bottleneck. Run static code analysis and automated tests on every change so reviewers spend their time on the decisions that need a person.
- Back up your org. If an AI-generated change deletes metadata or overwrites data, you need a recent backup you can restore from quickly. Among teams using AI, those with operate tooling are more than twice as likely to be confident in the security of AI-generated changes (43% against 20%).
- Get visibility after release. You need to see what AI-generated changes do once they’re in production, and catch Flow and Apex errors before your users do. Right now, 18% of teams find most of their issues in production. Observe has the strongest link to AI confidence of any stage: teams using AI with observe tooling are more than twice as likely to be confident in the security of AI-generated changes (45% against 20%). When you can see what a change does in production and act quickly when something breaks, you can trust a change wherever it came from.
Agentforce agents reach production where the pipeline already works
Building Agentforce agents is a different job from using AI to write changes, but it relies on the same pipeline. An agent is still a change to your org. It gets built in a sandbox, tested, released to production, and monitored once it’s live, the same as a Flow or an Apex class. So it makes sense that agent delivery tracks DevOps maturity closely.
The report asked teams where they sit on Salesforce’s Agentic Maturity Model. At Level 4, multi-agent orchestration, 100% of teams have tools or a process for every stage of the lifecycle. At Level 3, complex orchestration across multiple domains, 91% of teams do. Among teams not implementing AI in Salesforce at all, the figure drops to 70%.
Agent counts follow the same pattern. Across all respondents, 51% haven’t deployed a single agent. Teams with every stage covered are more likely to have deployed at least one (53% against 34%) and nearly three times as likely to have deployed three or more (25% against 9%).
Trusted AI and DevOps on one platform
Teams using Gearset get trusted AI changes from Cam, the Gearset Agent, and then deliver through their trusted DevOps pipeline - both sides solved on one platform. And the sum is greater than the parts. For example, Gearset’s observability solution helps you triage production issues such as Flow and Apex errors, and from there you can assign fixing the error to Cam.
Every change Cam makes goes through your DevOps pipeline with automated testing and code reviews, the same as your team’s work. You get full visibility and sign-off on what Cam is working on, plus an audit trail for compliance. That pairing answers the confidence gap in the survey: an agent that knows your org, with a lifecycle that checks its work at every stage.
The teams furthest ahead with AI built a lifecycle that catches mistakes, whoever makes them, and then let AI speed up the work inside it. If you’re ready to hand your backlog to an agent that works alongside you and delivers through your pipeline, book a demo of Gearset.
