At TDX 2026, Salesforce turned the entire platform into APIs, MCP tools, and CLI commands that agents can drive directly. Here's what's generally available today, what's still in preview, and how we're advising clients this week.
TDX 2026 wrapped yesterday in San Francisco, and the headline from day one was Headless 360. If you caught the keynote, you probably heard the word "agents" more times than you could count. If you skipped the stream, the short version is this: Salesforce spent the last year rebuilding almost every part of the platform so that agents, scripts, IDEs, or anything else that can speak an API or MCP can drive it directly. The UI is now one surface among many.
That is a bigger shift than most of the coverage is giving it credit for. Let's walk through it.
The 30-second version
Headless 360 turns the full Salesforce platform into callable services. Every major capability is exposed as one of three things:
- An API call
- An MCP tool (Model Context Protocol, the open standard most AI clients use to invoke external systems)
- A CLI command
That is it. The UI is still there, but it is no longer the privileged path into the platform. Agents, whether yours, Salesforce's, or a third party's, can build, deploy, query data, trigger flows, run tests, and ship changes without ever opening a browser.
Salesforce is shipping more than 60 new MCP tools and roughly 30 preconfigured coding skills in the first wave. That number is going to keep climbing.
Why this matters if you run a Salesforce org
We work with teams whose admins, architects, and developers spend a meaningful portion of their week in clicks: navigating Setup, running reports, moving metadata between sandboxes, configuring flows, opening tickets. Any one of those tasks is fast. The aggregate is expensive.
Headless 360 is a bet that the aggregate is expensive enough to justify putting a programmable interface in front of all of it. Once you can say "deploy this change set, run the tests, roll back if anything fails, and page me if the production org hits more than five validation errors," the operating model of a Salesforce team starts to look different. Fewer clicks, more orchestration. The shape of the work changes before any of the org charts do.
You don't need to buy into the full agentic vision to benefit. A competent engineer with Claude or ChatGPT open next to their IDE can now do things that used to require three consoles and a lot of patience.
The four pieces worth knowing
1. MCP tools and coding skills
The headline feature. Sixty plus tools give a coding agent live read and write access to an org: querying objects, running Apex, editing metadata, inspecting logs. The 30 preconfigured skills package common workflows, like "generate a flow for this process" or "scaffold a custom tab," so you are not prompting from a blank canvas every time.
For anyone who has been using an MCP client against Salesforce via community-built servers, this is a step up. Same protocol, official tools, proper scoping.
2. Agentforce Experience Layer
This one is easy to overlook. The Experience Layer separates what an agent does from how it is rendered. Instead of an agent replying with a wall of text, it can emit rich components: a flight status card, a rebooking workflow, a decision tile, a data grid. Those render natively inside Slack, Teams, Claude, ChatGPT, Gemini, or anything else that supports MCP apps.
The practical consequence is that you can build one set of agent logic and deliver it into the client your users already live in. No more picking between a chat experience and a portal experience.
3. Agentforce Vibes 2.0
Vibes is Salesforce's AI development partner. The 2.0 release adds full org awareness from the start, multi-model support (Claude Sonnet 4.5 is the default, with GPT-5 available), and two working modes: plan and act. It comes with a set of predefined agent skills for things like building custom tabs and generating flows.
If you tried the first version of Vibes, this is a much more honest product. It feels less like a chatbot pasted on top of Salesforce and more like a teammate who has actually read your org.
4. DevOps Center MCP
This is the one we are most excited about internally. DevOps Center MCP exposes the full CI/CD pipeline, so you can describe a deployment in natural language and let the agent handle the execution. The build loop that used to require context switching between four tools now runs inside one connected experience. Salesforce is quoting up to a 40 percent reduction in cycle time. We will want to verify that on our own projects, but even a fraction of that number is worth a lot on a complex release.
What is GA today, and what is coming
Generally available:
- Agentforce Vibes 2.0
- DevOps Center MCP
- Session Tracing
- Agentforce Experience Layer
Early access:
- Custom Scoring Evals
On the roadmap:
- Testing Center in May
- Salesforce Catalog in June
If you are planning work over the next two quarters, that sequence matters. The GA pieces are safe to build on. The May and June items are worth scoping around rather than into.
What we are watching
A few honest reservations before anyone starts rewriting their roadmap.
Governance. Giving an agent write access to a production org is not a thing you want to do casually. Salesforce has talked about Session Tracing and scoring evals as the guardrails. Those are the right ideas on paper. How well they hold up inside a real enterprise change management process is the next six months of testing.
The builder gap.The announcement assumes a certain comfort with APIs, MCP, and prompts. The admin-only team is not out of a job, but the line between "power admin" and "developer" is blurring faster than most career paths have adjusted for. Anyone who has been quietly investing in developer skills alongside their admin work is suddenly in a stronger position than they were last month. Teams that start that investment now will be in a better spot a year from here. Teams that wait will be explaining why their roadmap is slower than it used to be.
Pricing. Salesforce has not said much about how consumption gets measured once agents are doing meaningful work against an org. This is worth watching as GA features mature.
How we are advising clients this week
A few things we are telling teams right now.
- If you have a DevOps Center project in flight, start scoping the MCP integration. The ROI is concrete and the risk is low.
- If you have an in-flight Agentforce rollout, look at the Experience Layer before you ship any more custom UI. You may not need the UI you were about to build.
- Do not rewrite a portal or a Lightning app for headless today. The platform still runs the same way it did last week. Headless 360 is additive, not a migration.
- Spin up a sandbox, connect Claude or your MCP client of choice, and actually try it. An afternoon of hands-on will tell you more than any launch video.
Closing thought
The interesting question from TDX 2026 is not whether agents will do more of the work. That was already clear. The interesting question is whether Salesforce is the platform agents reach for. In a world where every major SaaS is becoming headless, the moat shifts from where the UI lives to where the data lives and what it costs to leave. Headless 360 is a credible bid on that future. Whether it ships the way the keynote implied is the next six months of watching.
If you want to talk about what this means for your org specifically, that is the kind of conversation we like having. You know where to find us.