Our Sponsors

Platinum
autorabit
Gold
enehano
ForVendi
PDF Butler
nextview
Silver
TITAN
SurveyVista
CGI
ADDA IT Solutions
Dragonfly 360
DemoJam Competitors
TITAN
SurveyVista
Nubes Tech
PDF Butler
Thank You
Plauti
Salesforce Trailblazer Community
CIO Applications
Nubes Tech
Robert Sösemann
Robert Sösemann
Director of Engineering & AI Lead at Aquiva

Agentic now!

How to make the best out of Agentforce

Agentforce now. Here’s why

  • Agentic is a the new norm. Software stops being automated — it becomes autonomous and conversational.
  • Salesforce will become Agentforce. Everything else is on hold. SaaS is dying. This is where the platform is heading.
  • It really works. With Aquiva, our customers moved past PoCs. They’re in production and seeing success.
  • Still simple today. AgentScript will add more complexity. Learn the basics while you can.
My Org Butler
In My Org Butler

Years of learnings, open source. github.com/aquivalabs/my-org-butler

1

Less often is more. Trust the LLM.

  • Fewer more powerful Actions. Less code more generic logic.
  • Using expensive AI mean learning to trust it (a bit)
  • Experiment to find a balance between too rigid and too crazy
  • Same with Topics . Find sweet spot between Overlap+Overload.
My Org Butler
In My Org Butler

ExploreOrgSchema + QueryRecordsWithSoql — 2 simple actions replace a complex dozens.

2

Supercharge actions with APIs

  • One API call is a function. A chain of them is a program.
  • REST API for data, Metadata API for config — your agent can touch everything.
  • It creates records, reads rules, deploys changes. No code, just conversation.
My Org Butler
In My Org Butler

CallRestApi + CallMetadataApi — read data, change config, all from a conversation.

3

More than record grounding: Data Lake

  • Not just Salesforce. Data Cloud ingests Snowflake, BigQuery, S3, files, web events — one place to ask.
  • Hybrid search beats SOQL. Keyword + semantic in one query — “Find deals similar to Acme” without a WHERE clause.
  • RAG on documents. Answer from your doc library, not from hallucinations.
  • Data Cloud Foundations is free. Start today.
My Org Butler
In My Org Butler

QueryDataCloud for hybrid search, AnswerFromDataLibrary + AnswerWithCurrentFile for files.

4

Ground in the world, not just your org

  • LLM training data is months old. Your customers ask about today.
  • Salesforce still has no built-in web search action. Tavily fills the gap.
  • Free tier, AI-optimized results, just a Named Credential away.
  • Combine with internal grounding — your agent knows your org and the world.
My Org Butler
In My Org Butler

SearchWeb — one Tavily API key, one Named Credential, done.

5

Let your agent learn with memory

  • Every conversation starts from scratch. Brilliant consultant with amnesia.
  • Persist preferences, decisions, tasks across sessions.
  • Memory turns a tool into a colleague.
  • Requires tricks to not be stopped as “Prompt Injection Attack”
My Org Butler
In My Org Butler

LoadCustomInstructions + StoreCustomInstruction — persistent preferences, zero code.

6

Headless agents work while you don’t

  • Scheduled, triggered, autonomous — the real value runs behind the scenes.
  • Humans in the loop only when needed.
  • Chat is how you demo it. Headless is how you ship it.
My Org Butler
In My Org Butler

HeadlessAgent + PlanForLater — “Every Monday, check stale opps and Slack me.”

7

Testing AI needs a mindshift 

  • Red/Green testing is turning into probabilistic evaluation
  • Testing Center has many flaws that eventually might be fixed
  • Useless Topic Match, irrelevant Action Match + Flaky LLM Judge
  • Use Testing Center for what it covers — then plug in custom tools for what it can’t: multi-turn flows, data resets, stronger judges.
My Org Butler
In My Org Butler

A Claude Code Skill generates multi-turn YAML tests, runs them through the Agent API, judged by your own LLM.

What I didn’t cover

  • Packaging. Agents in Managed Packages are painful — no working versioning, Agent Templates have no upgrade path.
  • Agentforce Analytics & Audit: How do people use it? How much does my agent cost? Where does it fail? How do people like it?
  • AgentScript: Migration won’t be simple. Deterministic reasoning — I’m still doubtful.
  • DevOps: How do you version agents, experiments and prompts? How do you deploy them? Deploy full of pain.
My Org Butler
In My Org Butler

Packaging, DevOps, analytics, security — it’s all worked through in the repo.

Robert Sösemann

Danke! Questions?

Robert Sösemann

Director of Engineering & AI Lead at Aquiva

/rsoesemann /rsoesemann + /aquivalabs