AI-driven Salesforce development has arrived in a big way. In-house, the company built their governance scaffolding, measurement system, and workflows first, while trying to get developers to adopt and adapt.

Late last year, they moved to Claude Code as their primary coding agent, rolled it out to everyone, and yanked all the token limits.

And now their developers are using AI to tear down and rebuild their own workflows. Work items per developer are up 50% over the prior year, pull requests merged up 79%, and their machine learning-based Effective Output score (value of code delivered, not necessarily quantity) is up 151% year-over-year.

For an example: When the company needed to migrate 33 API endpoints to a new cloud architecture, their traditional budget put it at 231 person-days (7 per API).

With Claude they did it in 13. Eighteen times faster. They’re also doing it with 100% test coverage and higher quality, too: increasing pull requests while seeing incidents drop 5%.

And here’s how this production is catching fire: These in-house Salesforce AI solutions are now helping revamp how all Salesforce developers can work.

They’ve provided tools (like Salesforce Skills) that are both accelerating and standardizing AI coding work for developers everywhere.

Let’s look at the details.

Why Salesforce still needs development

CRMs have always aspired to a business-friendly approach to data and workflows, and in Salesforce a lot can get accomplished just with Flow, validation rules, and Salesforce Setup.

But in the messy real world, business requirements typically push past this.

Maybe you have complex logic or custom calculations across records, need to connect Salesforce to your ERPs or middleware, have outside data warehouses or marketing platforms, want customized UI that doesn’t match Lightning components, or need more complex automation.

Custom development has always played an important role when the need went past Flow or configuration alone.

Salesforce’s Java-like Apex with SOQL opens this door, and the best Salesforce developers have always combined strengths in coding with knowledge of the platform and the business, too.

And of course, these needs haven’t changed, even when tapping AI for Salesforce development.

What’s changed is the how, the speed, and where expertise is most needed now.

How is Claude transforming Salesforce development?

In October 2025, the Salesforce-Anthropic partnership entered a new phase, with Claude

becoming a preferred model for Salesforce’s Agentforce platform.

At the same time, as mentioned above, the company adopted Claude Code globally for their in-house development.

Salesforce was an early Anthropic investor, but this move was about ensuring trust for regulated industries as well as getting in on the breakthroughs enjoyed by Claude Code’s fusion of frontier model and highly effective software.

(Salesforce termed this differentiation “code generation, customization, and data security.”)

Since, the relationship’s only continued expanding.

Claude’s Tag Slack integrations are both popular and powerful, in-house and out, with Anthropic creating 65% of their own product team’s code using an internal version of Claude Tag.

Claude Code and Skills underpin AI-powered Salesforce development

You don’t need Claude specifically to code in Apex syntax. But where the benefits are starting to show are through shared assets like the Salesforce Skills library, open source and maintained on GitHub.

Now when Salesforce developers work with Claude Code, they can take advantage of a

curated package’s specialized instructions across four steps:

1. Claude reads your natural language prompt and looks for the right skills, finding Salesforce-specific entries as needed.
2. The skills build what’s needed, drawing from the project for specifics like naming conventions and existing structure.
3. It reviews templates, can apply guardrails, and also generates test classes and required metadata files automatically.
4. Then it validates the code, checking for security violations and ensuring the required 75% test coverage.

This repository supplies much of what Salesforce developers need, with coverage for Apex, SOQL, Lightning Web Components, Flow, objects and fields, permission sets, Agentforce, and other parts of their platform.

MCP levels-up Claude for Salesforce development

Model Context Protocol (MCP) is Anthropic’s open standard that lets AI agents discover

and use tools that are approved and return the results when done.

With Salesforce, it’s applied through hosted MCP servers for data and business needs, Agentforce MCP connections for outside tools, and the Salesforce DX (Developer Experience) Server, also available via an open GitHub repository.

This offering combines with skills to enable agents the access they need to do more useful work. It reduces overhead and complication, and once again can radically accelerate the development loop.

With MCP, the agent doesn’t have to stop at each phase to go back to the developer with its code; it can interact with supported tools, execute its own tests, inspect failures, and loop again as needed.

The bottleneck shifts: What role does Claude play in modern Salesforce development?

Using Claude AI for Salesforce with these ready assets makes code generation faster and cheaper.

But as Salesforce’s Senior Director of Developer Relations René Winkelmeyer notes, this process doesn’t reduce the need for quality developers.

Instead, it shifts the importance from coders who can write the fastest to those with deep platform understanding and architecture capabilities.

Using AI, he notes, is also initially slower, as you have to rethink the entire SDLC.

It changes the kinds of errors you’re dealing with, and shifts the needed expertise to

design, quality engineering, and thinking across systems.

While the hardest part of the job (taking 90% of the time) used to be getting what you needed coded, AI-powered Salesforce implementation today shifts this into a breakdown of added design, time spent working with the AI, and the validation and editing of first draft code.

Software engineering has always sought to eliminate the repeatable, and today this capability is scaling up in a big way.

But it’s also shifting the importance of human responsibility.

Most AI-assisted Salesforce development is brownfield

Salesforce development doesn’t happen in an open space, whipping up wholesale

applications that will run in a vacuum.

In order to be useful, developers are working with Salesforce orgs that have already evolved and been customized, that operate in layers and require knowledge that’s not always cleanly documented and easily accessible.

This is where human expertise remains critical. Because while coding automation may craft a modification that’s well-coded and passes all needed tests, it can’t anticipate how that code will behave when fully interacting with the existing structure in real-world business use.

These systems can’t anticipate how things unfold (in their precise, but varied order), what workarounds have been put in place, and ultimately why decisions have been made the way they have.

As Winkelmeyer notes, with Salesforce development today, it’s necessary that “the architectural thinker unlocks everyone else’s productivity.”

Some hard limits to generative AI for Salesforce

It’s true and fair to say that many coding tasks are now being reduced from weeks or even months to hours.

Claude Code and the knowledge that developers are building and sharing around it (including for things like markdown files and maintaining some semblance of agent security) is making customization and adaptation far faster, provided the designs are right and serve a business best.

But even with engineered loops, AI systems can still bloat code and deliver the equivalent of slop, making it essential that developers re-purpose their workflows to know how and when to redesign and closely review the code.

It also doesn’t mean that complex tasks like wholesale enterprise Salesforce implementation can suddenly be accomplished in days.

Implementing the system goes far beyond code and is driven by need.

It requires understanding business processes, knowing what should be standardized, and what AI can and can’t customize easily (and painlessly).

The data remains foundational and directly related to quality, making things like cleaning and migration critical.

Integrations, permissions, and user testing all are driven by the real-world business at scale, and of course people need training for adoption and stakeholders need to be aligned.

Salesforce is an important system for many in regulated industries, but for real governance and security, change management and effective planning before and throughout implementation remains essential.

And while AI automation for Salesforce can be incredibly powerful, it also comes with a steep learning curve.

Because while AI can write familiar code, analyze data, communicate with other systems, and execute repeated steps faster than us in many cases, it can’t yet begin to understand the broader ramifications of many of its actions.

In other words, it doesn’t take responsibility, and it doesn’t keep an eye on the big picture

and the bottom line.

With AI, Salesforce consulting changes, too

All of this also applies to Salesforce partnership. The days of billing for long hours spent to write custom code are over.

The core value of a Salesforce implementation partner today is wholly in their ability to deliver successful, efficient outcomes.

For Salesforce consultants, it’s bringing an understanding of ROI, what components are best, and how the system can be applied overall to meet business goals. This means translating objectives to architecture, understanding the nuances of integrations and data dependences, and knowing how to carry the solution past deployment and on to widespread adoption.

It also requires understanding AI and the difference between speed and quality. Partners now must know what content can and cannot be safely exposed to AI agents, what development and security guardrails are critical, and how to really validate AI-generated work.

Peterson Technology Partners (PTP) is an example of one of these evolving companies.

They’ve provided tech consulting and recruiting for Fortune 500 companies for nearly thirty years, and have helped organizations get up and running with Salesforce for the past decade.

Their work includes both Salesforce consulting geared on ensuring businesses get the most from their Salesforce spend (whether new or existing), and Salesforce implementation services that perform all the hands-on digital work companies need.

Their AI expertise extends beyond Salesforce, but today they are regularly helping companies deliver safe, effective Agentforce automation that brings bottom-line results from AI.

Salesforce development automation has already changed the game

Salesforce’s success at delivering a 231-person days modernization project in 13 showcases a deeper transformation that’s well underway across the tech industry.

With more experience working with AI systems that are getting better—and software layers that bring together vetted skills, multiple-agent partnerships, engineering loops, external tools, testing, and real validation—this shift is becoming more durable, versatile, and far- reaching.

It most certainly changes the day-to-day experience of development, while shifting what skills are most important. Because the cheaper and faster code creation gets, the more important review, context, and architecture are becoming.

As an early investor in Anthropic and now a core partner, Salesforce clearly understands the future that AI plays in business as a whole.

But it’s still human expertise that knows what to build, how it fits, and most importantly

why it matters.

Frequently Asked Questions (FAQs)

How is generative AI changing Salesforce development?

GenAI is shifting where much of the Salesforce development time is spent. Practically, this means less time on manual coding in favor of agent-assisted execution, with more time on developing requirements, designing architecture, reviewing code, and establishing guardrails that are effective enough.

What are the benefits of using AI for Salesforce development?

Bottom line: AI increases developer productivity, accelerates testing and troubleshooting, reduces repetitive work, and improves documentation. But it requires an adjustment of workflows and has a steep learning curve. Ultimately, it’s now helping teams deliver Salesforce enhancements faster than ever.

How can Claude reduce Salesforce development time and costs?

Claude can now generate much of the Apex code and Lightning components, create tests, debug, document, and iterate in engineering loops. And while much Salesforce development work is brownfield and requires integrating with existing systems, the AI component can reduce manual coding effort and ultimately accelerate delivery.

How can a Salesforce consulting partner help businesses adopt AI for Salesforce?

Salesforce consulting partners can help identify the best, high-value AI use cases. They also can establish the right measurement metrics, prepare existing Salesforce environments, establish governance with security guardrails, integrate tools, and ultimately ensure faster development translates into lasting, safe business results.