Almost every Salesforce team is now using AI as part of their daily workflow. But many are missing out on the capacity gains they could see by reaching advanced AI maturity.
Ask a developer or admin how they use LLMs, and the answer is rarely that they have been able to meaningfully delegate work. More often, teams are using AI to speed up isolated tasks like explaining a validation rule, drafting an Apex trigger, or debugging a misbehaving flow. This is copilot-level AI maturity, not a virtual teammate.
In this article, we’ll take a look at what Salesforce teams are missing out on by using AI as a copilot rather than a virtual teammate, and practical steps you can take to reach that next stage of AI maturity.
Why Teams Aren’t Getting Past the ‘Copilot’ Stage
If you’re using AI just to speed up your own tasks, or are giving it user stories but have to do a lot of rework on the output it gives you, you’re using AI as a copilot. Lots of teams hold back from delegating more work to AI for two reasons:
- Techstack: Generic LLMs generate changes based on Salesforce documentation, not on how Salesforce teams actually work. So those changes often fail validation and are undeployable. Even when it produces technically valid changes, they don’t necessarily align with internal coding standards, security requirements, and your org’s existing metadata. Needing to carry out extensive rework to get a change deployment-ready isn’t what you’d expect from a colleague, so the AI isn’t working like a virtual teammate.
- Trust gap: If you’re the only line of defence between an AI-generated change and production, you’ll likely want to stay close to the change AI is working on to make sure it’s getting things right. But if you’re working within a DevOps process, you can trust that the deterministic guardrails of testing and security checks will catch any issues in the AI-generated change that you might miss during review – so you can do less handholding for your agent and delegate work to it more readily.
What Is a Virtual Teammate?
A more advanced AI setup doesn’t mean handing every decision to a fully autonomous agent. Most teams’ ideal state would sit between copilot and full autonomy: a virtual teammate. Rather than answering questions or suggesting code for a human to implement, a virtual teammate takes a ticket, can ask questions to clarify ambiguity, and returns a deployable, review-ready change that’s been planned, built, and tested against your own org. It’s a genuine shift in who does the work, not just a faster way to carry out your own work.
Can LLMs Alone Ever Be a Virtual Teammate?
Part of what keeps teams at the copilot stage is that their AI tool is really a collection of individual setups. Each developer or admin refines their own instance of AI with their prompts, configuration, and chat history over time. This holds you back from reaching a virtual teammate in two ways.
First, different team members could give the same ticket to their own AI instance and get meaningfully different results. That’s a hard thing to build a reliable process around because the team can’t reliably delegate work with certainty about the standard it’ll come back at.
Second, individual agent setups limit the value of coaching. When a virtual teammate is given feedback or corrections, that improvement carries forward the next time anyone on the team hands over a ticket – this makes delegating to the virtual teammate a genuinely shared investment rather than an individual one. A personal AI setup only ever gets better for the person who trained it.
Some teams are trying to tackle this with shared skills that the whole team use. That helps, but it requires upfront work to configure and is then a static file that only improves when someone directly edits and saves it. Feedback that’s then given by a user may be saved in their chat memory, but the correction doesn’t impact the shared skill.
The Virtual Teammate Checklist
Whether you’re evaluating AI tools or already have tooling in place, here’s a minimum checklist to validate whether your AI platform of choice will enable you to reach a virtual teammate setup:
- Org aware: You can give AI your repo context, but if a developer still has to hold the org’s specific quirks in their head and feed those details into their prompts, the tool hasn’t actually absorbed the org context. It’s still relying on a human to carry it. Without a genuine and up-to-date understanding of a specific org context, an agent will never work as an effective teammate.
- Consistent output: A virtual teammate can produce consistent standards and results regardless of who’s assigning the ticket, rather than being shaped by a private setup.
- Meets org-specific standards: A virtual teammate should be held to the same bar as any team member: understanding the org’s structure, meeting the quality standards already in place, and iterating based on feedback rather than needing its work reworked by other members of the team before it can be shared for review.
- Governed as standard: DevOps guardrails are what keep a production org secure as throughput scales, and every change should pass through the same testing regardless of who made it. If reviewing standards have to drop to keep up with an agent’s output, that’s a warning sign and should not be accepted as a necessary trade-off for increased velocity.
Introducing Cam: A Virtual Teammate Governed by DevOps
Gearset’s agent, Cam, is a virtual teammate that ramps instantly and can confidently tackle the tickets your team doesn’t have time for, so they can focus on strategic projects. Cam is automatically kept up to date with your org’s context so, when assigned a ticket, it plans the change using detailed context on the org’s metadata, builds the changes to the team’s coding standards, and pre-validates changes before proposing them for review – so you don’t waste time reviewing changes that won’t deploy successfully.
With built-in practitioner skills, Cam builds changes the way that admins and developers actually work in practice. And because Cam sits within your existing DevOps pipeline, every change goes through the same automated tests, security checks, and guardrails regardless of who authored them. So you can delegate to your virtual teammate knowing you’re staying secure and compliant.
“The rest of our company moves fast with general AI tools. We needed a way to move that fast inside Salesforce, without handing everyone read and write access to our production org. The Gearset Agent gave us the speed with the permissions and the governance intact.”
Marilyn John, Sr. Revenue Systems Manager, Hudl
Find Out How to Unlock Greater AI Capacity
Join the upcoming webinar, From Prompting to Delegating: How Salesforce Teams Unlock Real AI Capacity, on October 8 to get practical guidance on what’s actually limiting your team’s AI capacity today, what a true virtual teammate looks like in practice, and how to hand over a real ticket and get back a governed, reviewable change. The session includes a live demo of Cam picking up a user story and reducing a backlog in real time.
Can’t make it on the day? Register to receive the recording after the session has finished.
