Every year, Salesforce announces "the biggest transformation in its history". In 2026, for once, the phrase is not an exaggeration - but not for the reasons the marketing puts forward. Here is what we actually observe on the orgs we audit and deploy.
01AI has moved from demo to product
In our reading, Agentforce is no longer a keynote demo. Agents hold up in production on well-defined scopes - the very ones for which Salesforce ships standard subagents or templates: lead qualification and follow-up, support FAQ backed by a knowledge base, order tracking, meeting preparation. What has changed is less the language model than the tooling around it:
- Determinism before the LLM. The reasoning engine first runs a deterministic phase (variables, conditions, transitions); a filtered-out action is physically invisible to the model - you cannot "talk it into" performing it.
- Native testing tools: simulation without touching the data, batch test suites from a scenario file, session tracing. Tests run in a sandbox and consume billable AI usage - to be planned in the budget.
- Strict permissions: the agent acts through a dedicated user and its permission sets; it only sees what its rights allow.
- Governed versioning: draft, frozen version, activation - only one active version at a time.
The practical consequence: the question is no longer "do we need an agent?" but "which one, on which data, with which limits". The failures we see all share the same cause - an agent placed on an org whose data cannot keep up.
An AI agent is never better than the CRM it relies on.
02Data 360 has become the center of gravity
Data Cloud has been called Data 360 since October 2025, and Salesforce has made it the data foundation of its platform (now branded "Agentforce 360 Platform"): it is what feeds agent grounding, marketing segments and personalization, without copying the data. What that means for you: data debt has become strategic debt. The duplicates and empty fields that were a reporting irritant yesterday block your AI use cases today.
03Revenue Cloud: the convergence is settled
Salesforce has set the status of CPQ: end of sale, not end of life. Existing customers keep support and fixes, no forced migration has been announced, but product innovation is concentrated on Revenue Cloud Advanced - and new customers can no longer buy CPQ. If your org runs on a heavily customized legacy CPQ, 2026-2027 is the window to plan the transition - not to suffer it. We covered the subject in detail in From CPQ to Revenue Cloud.
04Licensing is being rethought around usage
Between Agentforce conversations billed per conversation or through flexible credits, and the "Agentforce 1" editions that bundle built-in AI and unified data, the cost model is becoming hybrid: per-user licenses and consumption, with a pool of credits shared between Data 360 and Agentforce. That is neither good nor bad - but it has to be managed, and the price lists move several times a year: always check the official pricing page of the day. In our experience, a license audit typically uncovers 15 to 25% of optimization headroom before any renegotiation.
05What does not change
- A clean data model beats any feature. It was true in 2016; it is even more so with AI.
- Adoption is won in the field, not in slides: training, visible quick wins, listening to users.
- Native first. Every custom development is a debt that comes back to haunt you at every release.
- Start small, deliver fast. The programs that succeed deliver value every 5 weeks, not every 18 months.
06Where to start in 2026
If you were to do only one thing this quarter: an honest assessment of your org - data, adoption, technical debt, AI readiness. That is exactly the purpose of our free Health Check: 30 minutes to know where you stand, and what deserves to be launched first.