Summer '26 hits production on June 12-13, 2026. The release notes weigh in at hundreds of pages — and like every Salesforce release, most of it is noise for anyone running a real org. The architect-relevant changes fit on one page.
This post is the filter. It assumes you've already skimmed the official notes (or had Trailhead summarize them) and want a strong opinion on what to actually touch in the next 30 days, what to wait on, and what's marketing without substance.
I'm not covering every feature. I'm covering the ones that change how you'd design a Salesforce project starting today.
Headline features that actually matter
Out of the dozens of items in the release notes, six change architectural decisions. The rest are quality-of-life or product-line specific. Here's the short list.
Apex default user mode + with sharing (GA, API v67+)
This is the most consequential platform change in years and almost nobody is talking about it. From API version 67 onwards:
- Database operations default to user mode instead of system mode
- Classes without an explicit sharing declaration default to
with sharinginstead ofwithout sharing
For greenfield code, this is correct and overdue. For brownfield orgs, it's a landmine — any class bumped to v67 that relied on the implicit without sharing to read records the running user couldn't see will silently start returning empty result sets.
Architect's takeaway:
- Don't bulk-upgrade existing classes to v67. Pin existing code to its current version. Only new classes ship with v67.
- Add an
ApiVersionESLint/PMD rule to your CI to prevent accidental upgrades during quick fixes. - This is also a chance to add an explicit
with sharing/without sharing/inherited sharingdeclaration to every class in the codebase. Make it a lint rule, not a code review concern.
Multi-Agent Orchestration (GA)
Agentforce can now coordinate multiple specialized agents on a single workflow, with shared context across channels. It looks magical in demos and is genuinely useful — but only if you've already solved the single-agent problem.
If your team doesn't yet have one Agentforce action in production with measured grounding accuracy and clear failure modes, multi-agent is a distraction. The orchestration layer multiplies every bad assumption you have about agent behavior. Two flaky agents talking to each other don't average out — they amplify.
Architect's takeaway: this is the right direction, but the entry ticket is real production telemetry on a single agent. Without it, multi-agent is a category of latency and cost problems you can't observe.
Flow Orchestration is now standard (GA)
Orchestration drops the usage-based metering and ships as a standard feature in Enterprise, Performance, Unlimited, and Developer editions. No more counting orchestration runs against a separate SKU.
This is bigger than it sounds. The previous pricing model made Orchestration commercially unattractive for use cases it was actually designed for — multi-step human approvals, async branching, retry-with-compensation. With the meter gone, the trade-off changes:
- Orchestration vs Apex queueable chains becomes a real architectural choice, not a budget conversation
- Long-running approval workflows that lived in custom objects + scheduled jobs can move to Orchestration
- "Saga"-style processes with explicit checkpoints are now affordable for medium-sized orgs
Architect's takeaway: revisit any multi-step workflow you wrote as a custom state machine in Apex. Some of those move cleanly to Orchestration now. Not all — Orchestration still has limits around step granularity and external callout patterns — but the cost/benefit math changed.
LWC State Manager (GA)
A first-party state manager for sibling and cousin component communication. No more event bus libraries from npm, no more bubbling CustomEvent through five layers of DOM, no more LMS hacks for components that don't share a parent boundary.
What you get:
- A central store you can read and write from any component on the same page
- Reactive subscriptions — components re-render when relevant state changes
- A documented API surface, which means it can show up in code reviews and PMD rules
Architect's takeaway: if you've been writing your own state primitives in LWC (almost everyone has), document the migration path. Don't rip-and-replace — pick one feature area, migrate it, measure the bundle-size and re-render delta, then expand. State Manager isn't free at runtime, and it's not a fit for every page.
LWC live preview (GA)
You can now preview a single component as you edit it, in browser or VS Code, without a full page reload. This is a dev-loop change, not an architecture change — but the architecture impact is real over time: shorter feedback cycles produce more componentized code. Teams that previously inlined logic to avoid reload pain will start factoring it out.
Architect's takeaway: update your component design guidance. Recommend smaller, single-purpose components and rely on the live preview tooling rather than monolithic page components.
Tableau MCP (GA)
Agents can query Tableau through the Model Context Protocol, with results flowing through the Agentforce Trust Layer. For orgs that put their dashboards in Tableau (most enterprises), this closes a meaningful gap — your agent can now answer "what was Q1 pipeline by region" without you exposing a custom Apex action that re-implements half the dashboard.
Architect's takeaway: if you've been writing Apex actions that wrap REST calls to your BI layer just so agents can read data, stop. The MCP path is the supported one. The catch: MCP support outside of Tableau is uneven — for Snowflake, Databricks, or in-house data lakes you're still writing custom actions, and that won't change this release.
What to deploy in 30 days
Low-risk, high-value items you can put in production before Winter '27.
Set with sharing explicitly on every Apex class
Independent of the API v67 default. Add a PMD or CodeAnalyzer rule that fails the build for any class without an explicit sharing declaration. This is a one-PR change with zero behavior risk and a clear security posture improvement. Do it before you start touching v67 code, not after.
Enable LWC live preview in your team's VS Code setup
Document it in your onboarding. Update your component design guidance. This is a quiet productivity win — measurable in PR cycle time, not in features shipped.
Migrate one Orchestration use case off Apex
Pick a multi-step approval or onboarding flow that currently lives as a state machine in custom objects + Process Builder/Flow + scheduled jobs. Rebuild it in Orchestration. You'll find the limits quickly — that's the point. Now you know the boundary for the next 5 candidates.
Replace email template ID references with names
Action Version 3.0.1+ in Flow now references templates by name, surviving deployments cleanly. If you've ever lost a release window to a broken email reference between sandbox and prod, this fixes the entire class. It's a search-and-replace migration with a clear payoff.
Turn on scheduled flow custom batch sizes
Scheduled flows now support batch sizes from 1 to 200 directly from the Start element. If you've been working around governor limits with multiple smaller scheduled flows or Apex wrappers, you can collapse them. Audit your scheduled flows, find the ones running over 50% of CPU/SOQL limits, and lower their batch size. Cheap reliability win.
What to wait on
Features that are real but need another release to stabilize, or have caveats the marketing doesn't lead with.
Multi-Agent Orchestration (in production)
The capability is GA. The operational tooling is not. You can build multi-agent flows today, but observability across agents is thin, cost attribution per agent is unclear, and the failure modes when one agent in the chain returns malformed output haven't been characterized by the wider community.
Wait one release. Build the single-agent foundation now — telemetry, eval harness, clear grounding scope — and let Winter '27 close the gaps in cross-agent debugging.
Ask Agentforce in Flow (Beta)
The promise — natural-language flow editing and runtime failure diagnosis — is compelling. The reality, in Beta, is that the generated changes need close review before save. The reasoning is sound for shallow flows; it breaks down on flows with > 10 nodes, nested subflows, or anything touching custom Apex actions.
Architect's call: useful for prototyping in sandbox, not for fixing production. Revisit at Winter '27 GA.
Elastic limits for async jobs (Beta)
The idea — async jobs exceed daily limits temporarily under load rather than failing outright — is genuinely useful for any team that's burned by Monday-morning batch spikes. But "elastic" needs a published policy. How much overage? For how long? What does the overage cost on your bill? Until Salesforce publishes those numbers, you can't design around it.
Wait. Don't reduce your existing rate-limiting until the overage contract is clear.
Setup with Agentforce (Beta)
Natural-language org configuration. Promising. Risky in any org with > 50 users. The same reasoning concerns as Ask Agentforce in Flow apply, plus the blast radius is bigger — a misinterpreted prompt against Setup can change permission sets in ways that don't surface immediately.
Sandbox-only for at least one more release. Even there, scope it to specific Setup areas (custom labels, custom metadata) and not the high-risk surfaces (permissions, sharing, profiles).
What's still mostly slideware
Features that got marquee billing but aren't yet usable in the way the announcement implies.
Agentforce Vibe IDE — for everyone
The IDE itself is real and ships in this release. The problem is the licensing change that ships alongside it: from June 1, 2026, non-Developer-Edition orgs pay for Agentforce Vibes. The free tier is Developer Edition only. For most enterprises this means the tool exists in marketing diagrams but isn't actually deployed on the laptops that ship production code. Until it's bundled with a SKU you already own, treat it as nice-to-have, not a platform standard.
FORMULA() in SOQL WHERE (Pilot)
SELECT Id FROM Opportunity WHERE FORMULA(Amount * Probability) > 10000 — this is a real productivity win when it ships. Right now it's Pilot, meaning you have to contact your Account Executive to even get access. Pilot features in Salesforce historically take 2-3 releases to reach GA, and the wire format on this one is non-trivial — query optimizer plans need to handle FORMULA() expressions sensibly.
Architect's call: don't refactor any code to anticipate this. Note it for the Winter '27 roadmap and move on.
"Process Compliance Navigator monitors live workflows"
The phrasing in the announcement suggests a generic compliance layer over any business process. The shipping product is narrower — it's part of Salesforce's Business Compliance Suite for regulated industries (Financial Services, Health & Life Sciences). The AI-assisted clause extraction from PDFs is genuinely useful for compliance teams in those verticals. For a generic Salesforce org with a custom approval flow, this is not the tool you're looking for.
Multi-agent orchestration across vendors
The announcement talks about agents working together as a unified team. The fine print: it's Agentforce agents talking to Agentforce agents. Cross-vendor multi-agent (Salesforce + OpenAI Assistants + LangGraph + whoever's next) requires MCP server bridges that are real for read-only data access but still hand-rolled for tool execution. The cross-vendor vision is closer than it was a year ago. It's not in this release.
How to think about the release
The pattern across Summer '26 is consistent: Salesforce is moving from "agentic features" to "agentic operations." Multi-Agent Orchestration, Tableau MCP, agents in Flow Builder — these aren't new ways to build agents, they're ways to run agents at scale. The plumbing matters more than the demos.
The Apex security defaults shift is the second pattern: the platform is closing decade-old footguns. User mode by default, with sharing by default — these are choices a thoughtful architect would have made manually in 2018. Salesforce is finally making them the path of least resistance. If your codebase relied on the old defaults, this release is going to surface that, eventually.
The dev-loop improvements (LWC live preview, State Manager, the Vibe IDE for those who can afford it) reduce friction in component-level work. They don't change architecture decisions on their own, but they change the cost of doing the right thing. Smaller components become cheaper. Componentization becomes the default, not the exception.
The honest summary: most of Summer '26 isn't urgent for any single architect. The two items I'd act on this month — explicit sharing declarations and one Orchestration migration — are both cleanup work that's been overdue for a while. The release just made them easier to justify.
Running Summer '26 in a sandbox already? Tell me what surprised you in the comments or on LinkedIn. I'm especially interested in the user-mode breakage stories — they're going to be the talk of Q3.
