The Uncomfortable Truth About Agentforce Vibes
- Sangamesh Gella
- Oct 21
- 3 min read
P.S. If you are reading this, you will definitely find it helpful. I write about Salesforce, AI tools, and productivity stuff that actually works: no fluff, no generic advice, just real experiences from the trenches. For more information, visit my website's home page below and subscribe if you haven't yet. Thank you for reading this.
Here is how it started. I spent 30 minutes configuring rules and context for Agentforce Vibes. I thought I was done. Then it asked me a question I couldn't answer.
Not "how do you spell this?" It asked me to clarify what constitutes a successful deployment. I am amazed because it prompted me on what counts as an edge case, whether I wanted to use generated URLs or custom string parsing. Five questions, maybe six. These are the questions a real developer on my team would ask before starting work.
I realised something was wrong with my thinking, not the tool.
Here's what Agentforce Vibes actually is: Salesforce's new coding agent, org-aware and specific to my metadata, my files, my patterns. It runs in VS Code or the browser IDE. It can read my codebase, understand my org structure, propose or execute changes. Available as an extension, though personal dev orgs are still waiting.
Think Before You Act
But the real tension lives in the Plan and Act modes.
The Plan doesn't do anything. It analyses. Proposes. Asks clarifying questions. I stay in control but feel like I'm moving slowly.
Act executes. Changes land. PRs get created. Branches get committed. It feels like real work is happening.
I noticed I wanted to skip Plan. It's uncomfortable. It exposes gaps.
When I pulled up a component in Apex, the agent didn't just suggest fixes. It identified missing imports, duplicate key logic, and edge cases around user ID handling. Then it asked: What's your confidence threshold? Do you want validation on payload shapes? How do you handle MIME type mismatches?
These weren't rhetorical. They were necessary. The agent was asking: "Are we aligned?"
That's when it hit me. I'm not hiring a developer. I'm building one. And that requires me to articulate what I actually want.
This is where rules come in. Not as restrictions. As codification. I write things like: "Write invocable Apex that can be called from flows when possible." "One trigger per object." "Use SF command for deployments." "Deploy only modified components, not the whole org."
These aren't rules the agent enforces on me. They're the rules I write down so the agent doesn't have to guess my taste. It's the difference between onboarding a junior developer and giving a contractor a handbook.
The Brain Behind All This Agentforce Vibes
MCP servers amplify this. Model Context Protocol lets me connect Agentforce to Jira, Linear, Confluence, and GitHub. It doesn't just work on code. It reads my tickets, creates branches based on issue descriptions, commits changes with context, and proposes PRs.
One workflow: Jira ticket > feature branch > code changes > commit > PR.
Except I'm still testing the connection.
Still figuring out how to chain Linear > changes > GitHub > deploy seamlessly.
The automation exists. The clarity about what should happen? Should it deploy immediately? Wait for review? — That's still being written.
That friction is real. And it's not a bug.
When I asked it to review code, it came back with: "Here are the gaps. Here's what might fail. Here's what I'd need to know to fix this properly." That transparency is the feature. Not the tool doing magic. The tool is honest about what it sees and what it doesn't understand yet.
But it hallucinates. It makes mistakes. I had to say it plainly: Don't rely entirely on the AI.
I still review every diff. I still think. The agent accelerates work I've already defined. It doesn't replace the definition.
So what's the real cost? Not money. Clarity. I have to know what I want before Agentforce can help me get it faster.
For teams with codebases where patterns are documented, processes where "deploy" has a definition, those teams will feel the acceleration immediately.
For teams still figuring it out, Agentforce becomes an honest mirror. It shows what I know and what I haven't articulated yet.
The paradox isn't that AI saves time. It's that AI saves time only after I spend time being clear. And maybe that's worth the 30 minutes up front.
You can watch more about it in real time below -
P.S. If you found this helpful, I write about Salesforce, AI tools, and productivity stuff that actually works: no fluff, no generic advice, just real experiences from the trenches. For more information, visit my website's home page below and subscribe. Thank you for reading this.
Comments