It's live. Summer '26 reached General Availability today, June 15, 2026 — after a week of daily posts building up to it. This is my first-impressions breakdown, written after spending the morning inside a live org. No press release rehash. Real observations.
The Headline Feature: Multi-Agent Orchestration Is GA
This week's series started here for a reason: Multi-Agent Orchestration is the structural centrepiece of Summer '26.
Powered by the Atlas Reasoning Engine 3.0, it works differently from what most people expect. There's no fixed decision tree. An orchestrator agent inspects registered subagents, reads their descriptions and available actions, and routes tasks dynamically. The agent decides who handles what based on what each subagent says it can do.
That's a genuinely different architectural model from what Agentforce shipped in Winter and Spring '26. And in a sandbox this week, it works as advertised.
The honest caveat: the quality of routing depends entirely on how well you've described each subagent. The description isn't documentation — it's a routing input. Vague descriptions produce unreliable routing. This is the thing most orgs will get wrong first.
My recommendation: don't deploy multi-agent orchestration in week one. Get your individual agents working cleanly first. Sequence orchestration into month two or three, once you've refined the descriptions that drive routing.
What Actually Impressed Me More: The Customer Engagement Agent
This one flew somewhat under the radar in the pre-launch coverage, and it shouldn't have.
The Customer Engagement Agent is a net-new capability that qualifies buyers autonomously, around the clock, across your website and email. Not a chatbot. Not a form with AI routing. An agent that converses with prospects, qualifies them against your ICP criteria, and hands warm leads to the right sales rep.
The gap it's designed to close is one of the oldest problems in B2B sales: a good lead comes in at 11pm, nobody follows up until 9am, and by then the prospect has talked to three competitors. This agent works while your team sleeps.
The setup isn't trivial — you need to define your qualification criteria, connect it to your CRM data, and tune the handoff logic carefully. But the concept translates directly into pipeline impact in a way that's easy to quantify.
The Most Underrated Addition: Momentum
Salesforce acquired Momentum in February 2026. It's now integrated into Summer '26, and it solves a problem that has frustrated every sales organisation I've worked with.
Momentum captures calls, emails, and meetings — across Zoom and Google Meet — and writes structured data back into Salesforce in real time. Call summaries, next steps, sentiment signals, deal context. All of it lands in Salesforce automatically, without reps changing a single thing about how they work.
If you've ever tried to enforce CRM hygiene through process, training, or management pressure, you know how hard that is. Momentum solves it at the source by removing the manual work entirely.
This is the feature I'll be recommending first to most sales org clients. The ROI case is clear, the friction is low, and the Agentforce context it feeds downstream makes every other AI feature more useful.
The Infrastructure Play Nobody's Talking About: MCP Servers GA
Salesforce-hosted MCP (Model Context Protocol) servers reached General Availability in Summer '26, and most of the coverage I've read is underselling what this means.
External AI agents — whether built on Claude, GPT, or any other foundation model — can now call into Salesforce natively. SObject CRUD operations, SOQL queries, Data 360 queries, Tableau analytics: all of these are now exposed as tool sources that external agents can consume.
This is the architecture decision that positions Salesforce as the hub of your AI stack, not just one node among many. If you're building any kind of agentic system that touches customer data, this changes the options available to you significantly.
Most teams won't touch this on launch day. But if you're thinking 12 months ahead about your AI architecture, this is the feature to understand now.
The Thing That Will Catch Teams Off Guard: API v67.0
This is the breaking change that didn't get enough attention in the pre-launch build-up.
With API v67.0 — effective across the June 13–15 deployment window — SOQL queries, SOSL, DML operations, and Database methods now default to user mode, not system mode. Apex classes now default to with sharing.
If your org has custom Apex written for earlier API versions, this can change behaviour in ways that aren't immediately obvious. Queries that previously returned records regardless of the running user's sharing rules may now return fewer records. Operations that previously bypassed field-level security may now respect it.
This is a security improvement — the new defaults are more correct. But legacy code that depended on the old defaults can behave unexpectedly after the upgrade.
What to do today: if your org has significant custom Apex, run a targeted audit of any code that queries across sharing boundaries or performs DML without explicit sharing declarations. Fix the code, don't work around the security improvement.
MeetTheMind Insight 💡
I've been in enough major Salesforce releases to know the pattern: the orgs that get value on day one are the ones that spent the two weeks before it in their sandbox. The orgs that are scrambling today are the ones that didn't.
Summer '26 is a solid release. The headline feature (Multi-Agent Orchestration) works but needs careful sequencing. The less-hyped features (Customer Engagement Agent, Momentum) have more immediate deployment value for most orgs. And the infrastructure shift (MCP servers GA) will matter more six months from now than it does today.
The best thing you can do this week: pick one feature, get it working properly, and measure it. Don't try to adopt everything. Depth beats breadth in every Salesforce implementation I've ever been part of — and this release is no exception.
Key Takeaways
- Multi-Agent Orchestration is GA and works — but sequence it into month 2-3 after your individual agents are solid
- Customer Engagement Agent closes the 24/7 lead qualification gap; strong ROI case for most sales orgs
- Momentum (conversation capture) solves CRM hygiene by removing manual logging entirely
- Salesforce-hosted MCP servers GA — the architecture that makes Salesforce your AI hub, not just one tool
- API v67.0: audit your custom Apex now if you haven't — SOQL, DML, and Apex sharing defaults have changed
- Pick one feature, implement it well, measure it. Don't try to adopt the whole release at once.