I've been spending a lot of time lately thinking about where Agentforce actually makes sense, not in the abstract "AI will transform everything" sense, but in the practical "what problem does this solve and who's going to buy it" sense.
Public sector client experience kept coming up.
And the more I dug in, the more I realized it's one of the most compelling use cases out there and also one of the hardest sells.
Government agencies aren't struggling because they're inefficient. A lot of them are dealing with a genuine staffing crisis. Hospitals, schools, emergency services, local councils, they're all being asked to do more with less. A caseworker managing benefits applications isn't lazy. They're buried. A 311 call center isn't slow because nobody cares. They're understaffed.
That's the context Agentforce for Public Sector walks into. And it matters, because it means the pitch isn't "here's a cool AI thing." It's closer to "here's breathing room."
Agentforce for Public Sector sits on top of the existing Public Sector Solutions data model, so it's not starting from scratch. It already understands the objects, the workflows, the regulatory context. That's not nothing.
The use cases that seem to resonate most aren't flashy. They're mundane in the best way:
The City of Kyle, Texas put it to the test on their 311 line. Out of 7,724 requests, only 911 needed a human. The agent handled the rest. That's not a demo, that's a city actually running on it.
Here's the thing though: even with numbers like that, public sector clients are slow to move. And honestly? I get it.
These aren't companies where a bad decision costs you a quarter. When a benefits agent makes the wrong call, someone might not get housing. When an eligibility check fails, a family might not get food assistance. The stakes are different, and the people in these agencies know it better than any vendor does.
There's also the institutional memory problem. A lot of these teams have been through technology rollouts that promised the world and delivered a mess. The skepticism isn't cynicism, it's scar tissue.
What I've noticed is that the agencies making progress aren't the ones who got the best demo. They're the ones who had a vendor that actually sat with them long enough to understand why they were nervous.
The technical side of Agentforce is genuinely solid. Agent Builder is approachable, the pre-built actions are useful, and the Einstein Trust Layer addresses a lot of the "but what does the AI actually do with our data" questions upfront. That's not usually where things fall apart.
Where things get hard is everything else.
Clients come in wanting "AI" but haven't thought about what done looks like. You need to get to outcomes before you touch configuration, what gets faster, by how much, measured how? Without that, you're building toward nothing.
Then there's the staff piece. The fear that agents are coming for jobs is real and it doesn't go away with a reassuring slide. It needs actual change management, conversations, not decks.
And data quality always surprises people. The moment you put an agent on top of real records, all the inconsistencies and gaps you've been living with quietly become very loud problems.
The approach that seems to work: pick one workflow, make it narrow, make it high-volume, and show results in 60-90 days. A 311 line. Permit status lookups. FAQ resolution. Something where the volume is real but the stakes of getting it wrong aren't catastrophic. Win there first, then expand.
I find this space genuinely interesting, not because the technology is impressive (it is, but that's not the point), but because the problems are real and the people on the other side of them actually need help. The opportunity isn't in dazzling agencies with capabilities. It's in earning enough trust to get started.
That's a harder thing to scale. But it's the actual job of an architect.
Burning tokens isn't a productivity metric, it's a cost center. Here's how the frugal AI shift translates into concrete architecture decisions on Salesforce projects.
Hooks let you enforce behavior in Claude Code instead of hoping the model does the right thing. The mental model, three Salesforce-specific hooks, and why this matters more in Salesforce than in most stacks.
If your architecture only lives inside the Salesforce landscape, your role is shrinking. Here is why modern architects must master enterprise patterns outside the CRM sandbox.