Dreamforce 2026 closed on September 17. If you were in San Francisco, you already know the week had a single organizing idea: Salesforce no longer wants the CRM screen to be the place work starts. If you watched from your desk, the same idea arrived as a flood of product names, including AIforce, Claudeforce, Slackforce, Koa, Coworker, Hunter, Casey, Fin, and a claim that the company now has 30,000 customers live on Agentforce.
This post is the briefing we promised before the event. Two weeks ago we published five predictions for nonprofit executives. Below we score those predictions, separate what is generally available from what is still in pilot, and translate the week into decisions a leadership team can actually make this quarter. The goal is not to recap every announcement. It is to help an executive director, a VP of development, or a chief operating officer decide what deserves a board conversation and what can wait.
The one-sentence version of the week
Salesforce spent Dreamforce arguing that your data, permissions, and business logic should travel into the tools your people already open, including Claude, Slack, and the Lightning search bar.
For nonprofit leaders, the implication is sharper than the keynote made it sound. Most of your staff do not live in Salesforce all day. Gift officers live in email and calendars. Program staff live in case notes, Slack/Teams, and shared drives. Finance lives in spreadsheets. If the platform can meet those people where they already work, without a new login and without a new training curriculum, Agentforce stops being a demo and starts being a staffing strategy.
What Salesforce announced
1. AIforce: the interface is no longer the product
AIforce is Salesforce’s new live interface layer. It takes the data, workflows, semantics, and security model you already have in Salesforce and exposes them to conversational surfaces. At launch, those surfaces are Claude (Claudeforce), Slack (Slackforce), and Agentforce Coworker inside Lightning.
Claudeforce is the one that will get the most attention in your next staff meeting. It is a plugin that signs into Salesforce as the individual user and only sees what that user’s permissions allow. It launched with 37 prebuilt sales skills, including meeting prep, deal health, and pipeline questions, and is in open beta for approved organizations. Slackforce is the more immediately useful surface for many nonprofit teams: Slackbot can pull live Salesforce context into a channel, and Slackforce Surfaces can generate a working dashboard, report, or calculator from a plain-language request.
Coworker is the conservative option. It sits in the Lightning search bar and acts as an AI teammate inside the product your power users already open. Salesforce said 100,000 users activated it in its first 35 days.
The Belmar Opinion: We are super excited for what this announcement will do for our customers. Conversational CRM is here, and we’re excited to start working with our clients to equip them with this powerful capability starting right away.
2. Koa: Salesforce built its own CRM brain
Koa is Salesforce’s first CRM reasoning model. It is post-trained from NVIDIA’s Nemotron family on synthetic CRM scenarios, including lead generation, opportunity qualification, multi-step service work, and it runs inference inside Salesforce’s trust boundary. It is in pilot now, with general availability expected in Winter 2026 for U.S. regions.
The strategic point is not that Salesforce suddenly wants to compete with Anthropic or OpenAI on general intelligence. The point is cost and grounding. Frontier models are expensive and generic. A smaller model trained on the actual work Agentforce agents do can handle routine CRM reasoning cheaply, then hand the hard cases to Claude or Gemini.
The Belmar Opinion: Our view is that Koa will be a game changer, but it’s going to take the next few quarters to see the benefits. Our recommendation is to keep an eye on the pricing and look for success stories that relate to your use case. If Salesforce can route everyday nonprofit tasks like donor FAQ, registration lookups, and grant deadline flags through a cheaper purpose-built model, consumption costs get less frightening. That is the story worth tracking.
3. Named agents and a longer attention span
Salesforce expanded Agentforce with named agents launching Casey for service, Paige for IT and HR, Carter for commerce, Hunter for outbound sales, Marshall for supply chain, Piper for inbound pipeline, and Fin from the Intercom acquisition. Hunter uses a long-horizon runtime, so an agent can pursue a goal over days or weeks. Most of those names map to commercial jobs.
The nonprofit equivalents were already in market or on stage: the Donor Support agent that lets a donor change a recurring gift without calling; the Participant Management agent that drafts case notes from Slack; America on Tech’s grant reporting agent, which cut report generation from four days to one hour; and Make-A-Wish UK’s Wish Matching agent. YMCA of San Diego remains the cleanest public proof point with 400,000 members, 24-hour multilingual support, and an anticipated 50 percent drop in membership administration. That is a specific bet: answering “how do I sign my kid up for swim lessons” at 9 p.m. is not the highest use of a staff member’s Tuesday.
4. Builder Central, Quick Pages, and the headless layer
Admins got Builder Central, a no-code workspace for apps and agents from natural language, in beta the week of the conference for Agentforce-enabled orgs. Quick Pages, expected in October, lets permissioned users build single-page apps on existing data. The Headless Experience Layer is now generally available: define business rules once, then render them as a Lightning component or a Slack message. The constraint in most nonprofit orgs is not imagination. It is an accidental admin with a growing backlog. Faster building is only a win if someone is still accountable for security, naming, and data quality.
5. The Trusted Enterprise AI Harness
The least photogenic announcement may be the most important one for a board. Salesforce bundled context, agency, actions, governance, security, and model management into what it is calling a Trusted Enterprise AI Harness, a control plane for discovering agents, assigning identity, setting policy, managing lifecycle, and watching cost across Salesforce agents and third-party ones.
If you only remember one architecture idea from the week, remember this: agents without a control plane become a second shadow IT problem. Nonprofits that skip this conversation will spend 2027 explaining to auditors which bot emailed which donor, with which data, under whose authority.
How our five predictions held up
On September 11, we published five predictions for nonprofit executives. Here’s how we did.
Prediction 1. Claudeforce expands past sales, but nonprofits do not get their own skill pack.
Result: Correct. Claudeforce launched as a platform capability with 37 sales skills and a clear path into Slack and other surfaces. There was no nonprofit skill pack on stage. A major gifts officer can still get value with prospect briefing, meeting prep, and a first draft of a stewardship note, because those tasks resemble sales work. A grants manager dealing with compliance language, reporting calendars, and restricted funds will not find a prebuilt skill that understands that logic. If you want that, you’ll have to build it.
Prediction 2. The next Agentforce Nonprofit agents target grants and program outcomes.
Result: Partially correct, and arriving from the customer side rather than a single product keynote. We did not get a splashy “Grantforce” launch. We did get customer examples pointing in this direction, including America on Tech’s grant reporting agent, Make-A-Wish UK’s matching agent, participant-management work, plus continued investment in Agentforce Nonprofit for fundraising, programs, outcomes, and volunteers. Summer ’26 already shipped the Donor Support agent, and the direction moving forward is grants and outcomes.
Prediction 3. Salesforce leads with nonprofit ROI numbers, and you should read them carefully.
Result: Correct. The nonprofit keynote and the impact sessions leaned on hours returned to mission. YMCA of San Diego’s anticipated 50 percent reduction in membership administration, which is amazing! Hours saved will direct more time to stewardship calls, program quality, and staff training.
Prediction 4. Some form of AI consumption relief for nonprofits.
Result: Not announced, at least not as a headline. The Power of Us program still donates 10 Enterprise licenses, but agent usage is still metered through Flex Credits and, in some cases, per-resolution or per-conversation models. Salesforce did announce simplified commercial packaging into Core, Advanced, and Max editions with bundled credit allotments just before the event, but more will be announced in the coming weeks.
Prediction 5. The NPSP conversation gets more direct.
Result: Salesforce doubled down on the fact that no additional investment is being made in NPSP. Every announcement made at Dreamforce in some way connects to Nonprofit Cloud (NPC). If your organization is running on NPSP, you need to be aware (if you’re not already) that the future is NPS. Does this mean that your system will stop working? No! But when all investment is going into NPS, it means that you’re losing out on innovation from Salesforce, and the longer you wait, the greater the gap that will form between what is possible.
A 90-day agenda for nonprofit leadership
You do not need a transformation office. You need four decisions and one pilot.
Decision 1. Pick one process, not a platform.
Choose a single, high-volume, low-judgment workflow that already lives in Salesforce and improve on it. Membership questions. Recurring-gift changes. Volunteer shift reminders. A first-pass grant report assembled from data you already trust. Write down the current cost in staff hours and work to improve.
Decision 2. Audit permissions before you connect Claude or Slack.
Claudeforce and Slackforce inherit your existing sharing model. That is the feature. It is also the risk. Have someone who understands role hierarchy, sharing rules, and field-level security walk a “what would Claude see if it signed in as this gift officer / this case manager / this intern” exercise. Fix the gaps before you connect.
Decision 3. Decide where saved hours will go.
Put it in writing. If the result of the above is saved time, where will you deploy those hours? Habit 2 from Dr. Stephen R. Covey’s famous book, The 7 Habits of Highly Effective People, is to “Begin with the end in mind”. Write it down and then execute!
Decision 4. Put NPSP on a dated decision, even if the answer is “not this fiscal year.”
The cost of staying is no longer theoretical. It is every agent, every headless experience, and every AIforce surface that will be built for Agentforce Nonprofit and not for NPSP. Can you continue to run on NPSP? The answer is, of course, yes! But there’s an opportunity cost tied to this decision.