The Big Bets
TDX 2026 (April 15–16, Moscone West): Headless 360 reframing, Agent Fabric for multi-vendor AI, Agentforce Vibes 2.0 multi-model developer assistant, Multi-framework React support. Consistent theme: Salesforce as platform, not destination. The keynote made the architectural shift explicit — every product surface (Sales Cloud, Service Cloud, Data 360, Tableau Next) gets exposed via MCP tools and via REST so agents and external apps can compose them. The closing demo showed a customer-built React app calling Salesforce records directly through MCP without a single line of LWC, with full FLS, sharing, and Trust Layer enforcement intact.
Developer Experience
Vibes 2.0 is the headliner. Schema-aware, multi-model, conversational coding. React as first-class development path. Faster inner loops with Summer ‘26 LWC preview. The Vibes 2.0 model menu defaults to Claude Sonnet 4.5 with one-click switches to GPT-5, Gemini 2.5, and Salesforce’s xGen-Code. Schema awareness means a prompt like “write an Apex trigger that prevents Opportunity close when no Quote exists” produces code that references your actual custom fields, not invented placeholders. The Salesforce DX MCP Server (Developer Preview) lets you run sf commands from Claude Desktop:
> deploy the Contact validation rule changes to my QA scratch org
[MCP] sf project deploy start --target-org qa-scratch --metadata ValidationRule:Contact.*Platform Direction
API-first, MCP-native, composable. Salesforce as capability layer. Agent Fabric for multi-vendor coordination. The platform becomes what developers build on, not a product they log into. Agent Fabric specifically positions Salesforce as the trusted control plane: third-party agents (Anthropic, OpenAI, Microsoft Copilot, internally-built) register as participants, and the Trust Layer enforces masking, audit logging, and policy across all of them. The deterministic orchestration primitives (Topic, Action, Plan) give you finite-state guarantees that pure-LLM orchestration can’t.
Planning Priorities
Familiarize teams with Headless 360 and MCP. Pilot Agent Fabric for multi-vendor workflows. Evaluate Vibes 2.0 for developer productivity. React projects become viable — pick your first candidate. The realistic 90-day plan most architects sketched after TDX:
- Days 1–30: Spin up a sandbox, install the DX MCP Server, pilot Vibes 2.0 with two developers
- Days 31–60: Identify one external surface (Slack, voice IVR, partner portal) and prototype as a Headless 360 client
- Days 61–90: Score Agent Fabric against your incumbent agent vendor lock-in; build the migration path even if you don’t execute
What Changed in 2026
The Salesforce of 2024 sold Lightning pages and admin clicks. The Salesforce of 2026 sells governed APIs, agent runtimes, and a metadata layer that any framework can consume. Lightning isn’t deprecated — it’s now one consumer of platform capabilities, alongside Slack, custom React apps, voice channels, and external agents. Practitioners who specialized in click-config will need to add API and MCP literacy; pure declarative skills cap out at smaller orgs.
Common Failure Modes
Treating TDX announcements as production-ready. Most of what was announced is Pilot, Beta, or Developer Preview. Headless 360’s experience layer is in Pilot for select customers only. Agent Fabric’s multi-vendor coordination is announced but not GA until Winter ‘26 at the earliest. Don’t refactor production systems against pre-GA timelines. Second failure: assuming React multi-framework eliminates LWC. Record pages, utility bars, AppExchange components, and any deeply embedded Lightning surface still want LWC for the foreseeable future.
What to do this week
Watch the TDX 2026 keynote and the Agent Fabric deep-dive session on Salesforce+. Install the Salesforce DX MCP Server in a developer scratch org and run one task end-to-end through Claude Desktop. Score your team’s MCP/API readiness honestly so the next quarter’s roadmap reflects actual capability.
