Ultimate guide

to creating
AGENTS By Ian Gotts, Adrian King, and Jack Lavous

Ultimate Guide to Creating Agents - Elements.cloud 1

Executive Summary
The excitement since the launch of ChatGPT has been mixed with skepticism about the occasional
impact of hallucinations. However, the gains when used in specific use cases are huge. Elements has
embraced a culture of AI experimentation and we are seeing this in every department. And we’ve
embedded AI into the Elements platform to transform the productivity gains for our users.

Agents move the AI game on again. There is the inevitable hype that they will disrupt every industry
and displace workers. Some of this is misplaced. However, their ability to take on repetitive, boring,
data-intensive work is amazing. Agents are more than a bot that has pre-defined decision trees. An
agent can take a natural language input and then, based on its instructions and prescriptive actions,
deliver an outcome. It can plan, reason, and execute actions. And based on its guardrails, hand a task
that exceeds its capabilities back to a human.

The use cases could be internal such as providing answers to HR policy questions based on
employees’ location, providing Sales SDR support for an account executive, giving support agents a
360-degree view of an inbound customer call. Or they could be customer-facing, for example, an
agent could handle product returns. By understanding the product, customer and related policies could
process a return, exchange, or warranty claim. And the power is that the user is asking questions using
their language.

Great agents are a competitive advantage. Poor agents destroy brand trust. There is a risk that you
are simply accelerating a poor process: your mess for less. Customers will gravitate to organizations
that have great support agents that are easy to deal with. Employees will want to work in organizations
where agents reduce the “busy-work”.

Building agents has become easier. Agentforce has enabled any Agentblazer to build agents using
the familiar drag/drop paradigm. Add in the ability to connect data from 3rd parties via Data Cloud
and there is the potential to transform user experience. The Agentforce metadata is straightforward
- Agents have collections of Topics - called Skills; Topics have Instructions and Actions; Actions are
executed by Apex, Flow, and the new Prompt Templates.

Building agents for the enterprise at scale is like any other tech-led transformation. It requires a proven
repeatable approach. This enables agents to be delivered quickly, meet the true business needs, and
have the required level of governance. Remember, agents are your brand ambassadors.

We’ve taken our experience from implementing GenAI since ChatGPT launched 2 years ago, 20+ years
of Salesforce implementations, rolling out Data Cloud internally, and now building agents. We developed
an implementation approach and highlighted what we believe are the key success factors:

Ultimate Guide to Creating Agents - Elements.cloud 1

 Go slow, to go fast: Agentforce will deliver results faster if you have a mature implementation
approach; process-led change, data governance, and metadata management. Use the first
Agent to put these in place to build a strong foundation. The organizations that have these
disciplines mastered will be able to deliver Agents more quickly. They will accelerate ahead of
their competitors. So, even if your organization does not feel ready to implement an Agent, then
now is the time to start working on these best practices.

It all starts with process: The initial agents are replacing or supporting business processes
that are currently delivered by humans. Use a UPN process diagram to get agreement on
those processes down to a level of detail that enables you to automatically build agent using
instructions and actions. If you skip the design step it will dramatically increase the time to
iterate to an acceptable agent, which will be frustrating, knock your confidence, and lose
executive support.

Start small. Think Big: Building agents will be iterative. The initial Agent is deployed with just one
Topic. You can then start extending the capabilities by adding actions to Topics, and Topics to
Agents. Focus on narrowly scoped use cases to build experience. The learning you get from this
first Agent is as important as the ROI.

Avoid DC complexity: Data Cloud is required for every Agentforce implementation because
that is where the tracking data is stored. To access data from external systems, or to query
unstructured data you need to implement Data Cloud. More complex use cases will slow down
the implementation of your Agent. This will rob you of valuable momentum and learning.

Agents can help build agents: We’ve designed a repeatable implementation cycle with
documentation templates. This standardization has enabled us to build AI support that can
accelerate the implementation cycle. This dramatically reduces the time to build an Agent.
However your knowledge of your business, your needs, and your expected outcomes will be
required to update the results.

Tooling: With every transformation change at scale, you need a platform to help manage this.
For building, testing, and running Agents that is Agentforce. For everything else - planning,
design, versioning, metadata impact assessment - you need Elements.cloud.

Those organizations that are best placed are already in good shape; well-understood processes, strong
data governance, and effective metadata management. But they need to lean-in and explore the art of
the possible. No organization can afford to use compliance, data quality, or risk as an excuse to sit on
the sidelines. The future winners have already started; they are piloting, experimenting and learning.
And they are accelerating away from the pack who may never catch up.

Ultimate Guide to Creating Agents - Elements.cloud 2

Table of Contents
Executive Summary 1
Authors 4
Who should read this 5
The big disruption 7
AI is here to stay 7
Enter the AI Agent 8
The end of humans? 9
JTBD - Jobs To Be Done 9
Your mess for less 10
Are you ready? 11
What is Agentforce? 12
Strategic. Integrated. The future. 12
5 Key Attributes of an AI Agent 14
How Agentforce works 16
Agent 17
Skills 18
Topics 18
Actions 19
Instructions 19
Metadata 20
Implementation approach 21
Repeatable. Proven. Flexible. 21
Set up & Manage Agentforce 22
Identify Agent Skills / Use Cases & Jobs To Be Done 23
Design JTBD 24
Build Topic 26
Test Topic 26
Deploy Agent 27
Operate & Monitor 29
Success secrets 30
Go slow to go fast 30
It all starts with process. Shift Left. 30
Start small. Think big. 31
Avoid Data Cloud complexity 31
Agents can help build agents 32
Using Elements.cloud 33
Final word 34

Ultimate Guide to Creating Agents - Elements.cloud 3

Authors
Ian Gotts
Ian is Elements.cloud co-founder and CEO and is best summed up with
3 numbers:

10: He’s a passionate educator and has written 10 books on Salesforce,

business analysis, change management and compliance.

20: He’s been a Salesforce customer for over 20 years and was a
speaker at the first UK “Dreamforce” which was Marc, 3 customers and
120 delegates in 2001!

30: For over 30 years he’s been a business analyst and he drew his first
process map working for Accenture in 1986. Since then he’s been talking
about the importance of business analysis at events around the world.

Adrian King
Adrian, Elements.cloud co-founder, COO, and CTO, has a robust
background in Research and Development. His earlier career involved
overseeing R&D at various global software firms, showcasing his depth
in the field. Previously, he served as the CIO and COO of Nimbus
and following this tenure, Adrian took on the role of turnaround CEO
for a startup in the oil and gas drilling industry. In this position, he
revolutionized their product strategy to align with the buying culture of
their target market. Known as an experienced software entrepreneur,
Adrian specializes in offering strategy consulting to numerous
businesses, focusing on IT leverage and operational optimization.

Jack Lavous
Jack, VP Business Excellence and Operations for Elements.cloud is an
authority in Salesforce Org management and stakeholder engagement,
Jack is crucial in ensuring Elements.cloud stays on top of its game. With
a keen eye for detail and a passion for streamlining processes, Jack is
dedicated to enhancing business performance and driving excellence.
Jack leads initiatives that optimize workflows and in doing so, maintains
the company’s position as the pinnacle of its industry. Beyond his role
as VP, Jack is an avid F1 enthusiast, with a love for travel.

Ultimate Guide to Creating Agents - Elements.cloud 4

Who should read this
The promise of AI to transform performance
has not yet materialized. Agents seem to be
another wave of hype. Whilst it seems prudent
to sit on the sidelines and wait until there is more
market maturity, we believe that this is a mistake.
Organizations that put in place the structure and
tooling to deliver agents at scale will have a huge
competitive advantage.

Building agents does not require deep AI skills. It

is a process-led, business transformation project.
Executives / Platform Owners It requires executive sponsorship, leadership,
and proven, practical, scalable implementation
approach. This book is a blueprint for success.

Business Analysts, Admins and Developers can

all become Agentblazers. The Agent Builder
tooling follows the familiar low-code approach,
and agents are powered by core metadata -
Apex, Flow and Prompt Templates - that every
Trailblazer understands.

They understand an agent cannot be built in

minutes that delivers real value for their org. But
how long could or should it take? And what needs
to be in place to be successful?

This book sets out an approach that Salesforce

professionals can use to deliver agents more
quickly and more consistently. With a proven,
Salesforce Professionals
practical, scalable approach it is possible to get
buy-in and support from executives who are
concerned about the risks of AI and exposure of
consumption pricing.

Ultimate Guide to Creating Agents - Elements.cloud 5

 Your clients are looking to you for advice on when
to engage with AI, how to mitigate risks, and
how to deliver agents that deliver competitive
advantage. But clients have heard that agents can
be built quickly and are expecting you to be able
to deliver on that promise.

This book sets out an approach that Salesforce

consultants can use to deliver agents more
quickly, consistently, and still defend your
margins. It can allow you to protect your industry
IP by capturing it in agent templates, which
Salesforce Consultants also creates strong competitive differentiation
from other firms who are simply jumping on the
Agentforce bandwagon.

Ultimate Guide to Creating Agents - Elements.cloud 6

The big disruption
AI is here to stay

The rapid advancements in AI models and the art of

prompting have revolutionized what AI can accomplish.
The benefits of a well-planned AI Agent implementation?
Monumental.

McKinsey’s latest report on generative AI’s economic

potential paints a clear picture of the transformation that’s
already underway.

But since the launch of ChatGPT nearly 2 years ago, AI

has followed a classic “hype cycle”. Initial expectations
were off the charts when ChatGPT first launched. But
with the early limitations, hallucinations, and perceptions
of immaturity, many people rapidly fell into what Gartner
defined as the “trough of disillusionment”.

Because of this, many business leaders are either in denial or “waiting to see what matures”. We believe
that is a profound mistake. Now is the time for business leaders to encourage their teams to experiment,
learn, and make mistakes. AI is improving exponentially. It is improving multi-modally, i.e. not just making
leaps in understanding and generating text, but voice, video and above all reasoning. Staying on top of it
is hard. Playing catch-up may be impossible.

At Elements.cloud, we’ve been using ChatGPT since the start.

It’s now deeply integrated across every corner of our business
— from product management to marketing, business excellence,
IT, and even finance. And we’re not stopping there; we’ve
embedded AI in multiple areas of our Elements.cloud platform.

What’s been fascinating is the cultural shift that’s emerged. Instead of fearing AI or being
dismissive of its capabilities, our teams have embraced it with a mindset of, “Why not?” or “What
if we tried?” Every task is now seen through the lens of “can AI assist with this?” The result? A
workplace that’s more productive, curious, experimental, and, quite frankly, more open-minded.

Ultimate Guide to Creating Agents - Elements.cloud 7

Enter the AI Agent

You’ve probably encountered those bots on websites that try (often in vain) to answer your simplest
questions. These bots operate on predefined decision trees, trying to guide you to the right resource.
The result? Experiences that range from “mildly helpful” to “press 1 for confusion, press 2 to get stuck
in a loop.”

The game-changer has been AI’s ability to inject a more conversational tone into these interactions. AI
can now interpret natural language inputs, but bots are still bound by their rigid decision trees. So, the
customer experience is still largely defined by the quality of the bot’s design.

As an example, at Elements.cloud, we use Intercom.com for our support

site — support.elements.cloud. We’ve got a huge library of resources
ranging from solution guides to deep technical dives. The old chat system
was decent; it could suggest relevant articles, but only if users knew
what they were looking for. Enter Intercom’s Fin, powered by AI. Now,
users can ask questions in natural language, and Fin interprets those
questions, locating the best resources. The result? Fin now successfully
resolves 78% of our support queries on its own.
78%

But Fin isn’t a true AI Agent—at least, not in the academic sense. Intercom has optimized it to refine each
query, fine-tune every response, and ensure high-quality answers. However, a real AI Agent can reason,
plan actions, and execute them autonomously. The key difference here is in the reasoning.

An AI Agent is equipped with resources—data, automated workflows, and even guardrails that tell
it when to pass things off to a human. Picture it: an AI Agent that can verify customer details, pull
order references, and deliver personalized responses. It can even take actions, like firing off emails,
rescheduling deliveries, or updating orders. Essentially, it’s like having a junior employee who can execute
parts of your business processes using the automation you’ve already set up.

But here’s where it gets interesting: we need to think of an AI Agent as that eager, fresh-faced new hire1
who’s just joined your team. Yes, they need to be taught about your organization, its culture, policies,
and the tools at their disposal. But let’s not forget—they’re also your brand ambassador, interacting
with thousands of customers. If they’re not on-brand, the cost-cutting pursuit of automating customer
interactions could backfire spectacularly.

So, your AI agents are not just about reducing costs; it’s about delivering an experience that aligns with
who you are as a company.

1
Probably a male 18 year old intern. Not always right, but confident, and never in doubt. 8
The end of humans?

Yes, in some cases. A large-scale assessment conducted by the

Institute for Public Policy Research (IPPR) on 22,000 tasks within 3,000,000 customer
the UK economy revealed that approximately 11% of tasks are service representatives
currently exposed to generative AI. This could expand fivefold if AI employed in 2022
systems are more deeply embedded into organizational processes.

In 2022, nearly three million people were employed as customer 66

service representatives, with the majority (66%) being women,
en

%
according to Data USA. These are the roles most at risk as AI

34 %

w o m en
agents become more prevalent. However, it’s not all doom and
gloom. No customer wants to endure a 30-40 minute wait to
resolve a simple issue that an AI Agent could handle instantly.
And let’s be honest, after that wait, customers aren’t exactly in the
best frame of mind when they finally speak to a human. AI agents
can manage lower-level tasks, liberating humans to focus on
higher-level, more rewarding work.

Working in a cubicle, rescheduling appointments, or answering basic queries might pay the bills, but
it’s far from fulfilling. The turnover rate in call centers is notoriously high because employees often
leave if they find a job that pays just a little better per hour. This churn is bad for individuals, bad for
organizations, and, ultimately, bad for customer experience.

JTBD - Jobs To Be Done

Initially, AI Agents will handle the simplest

“Implementing agents transactional tasks. A critical element of their
is much like any other success will be the seamless handoff to a
human when necessary. Over time, the best AI
technology-enabled business
implementations will set new standards for what
transformation. Time spent customers expect from their interactions. The
planning will help eliminate quality of customer service isn’t determined by
painful rework and get you a the skills or salary of the person on the other
end of the line. It’s about the effectiveness of
working agent far faster.”
the AI Agent—how well it’s been instructed, how
well it understands processes, and how clean
Jack Lavous,
the data is.
VP Business Excellence and
Operations, Elements.cloud
AI agents are poised to become a significant
competitive differentiator. A poor AI experience
can easily push customers toward a competitor
that has mastered AI Agent implementation.

Ultimate Guide to Creating Agents - Elements.cloud 9

Your mess for less

When brainstorming use cases for agents, the first one should be the simplest and easiest to build
confidence and momentum. As you gain experience you can take a more strategic and innovative
approach. Look beyond the desire to make any process faster and cheaper. Great agents are competitive
advantage. Poor agents destroy brand trust. There is a risk that you are simply accelerating a poor
process: your mess for less.

This is an opportunity to completely rethink your business and design an exceptional process based on
your organization’s brand values. Below are some different perspectives that will help you think outside
the box:

What is an exceptional user experience: for customers, partners, and employees?

Can agents make humans as effective as your top employee? Standing on the shoulders of
giants. Giving the most junior team member the smarts of your best and brightest employee.

How can an agent coach your teams?

What about “Slow AI”. Rather than a rapid, light touch on-line experience, is there an
opportunity to provide service that is more considered, in-depth, and researched and is
delivered async? How would that look?

Could an agent spot trends and alert customers with recommendations?

What could you never do before because you couldn’t aggregate the data, or every customer
interaction would require custom code to provide the answer?

Can you put the customer in charge: give them the agent so they are in the driving seat?

Organizations that are innovative and take an AI-first approach to redesign processes will be the future
winners. This requires out-of-the-box thinking. And don’t be disheartened if AI cannot deliver the solution
at the moment. We’ve been trying to get AI to draw process diagrams from a sketch for 5 years. Our last
attempt 12 months ago got close, but it was not good enough. It was GA in Elements.cloud last month.

Ultimate Guide to Creating Agents - Elements.cloud 10

Are you ready?

Organizations that can exploit agents will accelerate away

from their competition. Remember the huge gains the
stores made because they were able to move rapidly from
in-person to collection and/or delivery during COVID. Best
Buy implemented contactless curbside pickup that proved
242% increase in sales from
highly successful, with online sales growing 242%.
contactless curbside pickup
Best Buy
We see that organizations that are well-placed to maximize
agents have mastered the following:

Well-defined business processes: Agents are “virtual employees”. As you define the
processes that the agent will deliver, and the integrations back into the processes delivered
by a human, you should streamline them to deliver an exceptional user experience. Having a
strong culture of processes and business analysis is critical, and existing processes in place.

Strong data governance: Poor data is junk food for AI. An agent is working autonomously doesn’t
have the gut-feel of a human who may question data that seems off. An agent will simply
take action. You need to have confidence that the end to end data flow is transparent and
controlled. Data quality was always important. Now it is critical. Again, this is a core discipline.

Comprehensive application metadata management: AI needs to understand which application

functionality it can use to plan and deliver the results. Agents can reuse functionality that
already exists to drive the business: e.g. validate customer, or reschedule delivery. There needs
to be a rigorous implementation cycle that includes dependency and impact analysis. No one
should be making changes without understanding if it could affect an agent’s behavior.

Organizations with these core disciplines in place will see agents as another way to deliver better
services, more profitably. Customers will want to deal with organizations that have empathetic, accurate,
and effective agents that reduce CES (Customer Effort Score). In the war for top talent, employees will
want to work in organizations where agents can provide support for their role, and help them navigate
corporate policies and processes. Agents can fill the labor gap.

We see organizations that run on “employee heroics” because of a lack of clearly defined processes. This
leads to poor data quality, and is compounded by applications with poor adoption, that have not been
well documented. The root cause is often an implementation lifecycle that does not focus on the value of
planning and business analysis.

Organizations that are not in great shape should use the promise of agents as the catalyst to get their
house in order. Without these core disciplines that put in place the foundations, you can deploy agents at
scale with any confidence that they will not be a reputational, compliance and operational nightmare.

Ultimate Guide to Creating Agents - Elements.cloud 11

What is Agentforce?
Strategic. Integrated. The future.

Salesforce launched Agentforce which encompasses all the AI capabilities that can be delivered
through an agent that has access to your org metadata and data from external systems via Data Cloud.
The diagram below shows what Salesforce has put in place strategically.

It has built on the last 25 years of investment in the core platform meaning that every Trailblazer can
be a part of the new agentic revolution. You can get the background on Agentforce from the Salesforce
website or by attending one of the Agentforce World Tours or hands-on workshops.

Agents have the ability to reason, orchestrate complex tasks, and take action, delivering personalized
experiences at scale. By merging language processing with metadata-driven apps, they are fundamentally
changing not only how businesses operate but also how software is designed and deployed.

Ultimate Guide to Creating Agents - Elements.cloud 12

Agentforce agents are at the forefront of this transformation, setting the bar with their unique,
game-changing capabilities:

Trusted. Powerful.
Data protection is paramount. Agentforce These agents harness the full power
leverages the Einstein Trust Layer, of Salesforce’s platform to deliver
applying the same robust metadata, transformative experiences across
permissions, and sharing models that sales, service, commerce, marketing,
Salesforce applications have relied on and specialized industries. It’s not just
for years. Your data is safe, secure, and automation; it’s a reimagined customer
handled with the highest standards. experience.

Grounded in data.  Low-code tools.

By anchoring AI in the vast, unified Building and managing agents has never
datasets available through Data Cloud, been easier. With low-code tools like
Agentforce agents produce more Agent Builder, Prompt Builder, Model
accurate, relevant, and actionable Builder, and Flow Builder, you can design,
insights. This ensures every interaction customize, test, and deploy agents
is informed by the most complete and quickly—no heavy coding required. It’s
up-to-date information. the future of agility and speed in
AI-driven solutions.

Ultimate Guide to Creating Agents - Elements.cloud 13

5 Key Attributes of an AI Agent
An AI Agent’s ability to deliver exceptional customer outcomes is based on the quality of data it can
access, and the clarity of its instructions. Here are the 5 attributes that define a successful AI Agent:

Role: Defining the Scope and Instructions

It all starts with defining the tasks an AI Agent can handle effectively. This requires
a clear, detailed view of the process and outcomes. The agent is a digital employee
that needs to be instructed just like a human, except they have no context so need
more explicit instructions. The fastest way to understand those tasks in the context
of the desired experiences is to map out the operational processes. By mapping out
these processes, you can enhance them to leverage the Agent’s reasoning capabilities,
crafting a truly remarkable customer experience. The power of an AI Agent lies in its
ability to reason and take action—but that only works if its role is well-defined.
The scope of what it can and cannot do must be crystal clear because, as they say,
“The devil is in the detail.”

Data: Accessing the Right Knowledge

Data is the fuel that powers AI agents. However, in many organizations, customer
data is scattered across different systems. Humans can piece together a customer’s
interactions to provide a seamless service, but an AI Agent needs that 360-degree
view to be available as a search. The better the data quality and accessibility, the
more effective the Agent will be. This is where Data Cloud comes into play, providing
Agentforce with a unified, harmonized view of customer data, enabling more accurate
and relevant insights.

Actions: The Agent’s Toolbox

An AI Agent isn’t just sitting around waiting for instructions—it’s making decisions and
taking action. These could be automated workflows, prompts, or integrations with
external services. Whether using existing metadata or newly created resources, the
Agent’s effectiveness depends on the actions it can execute based on the data and the
task at hand. However, if your applications are burdened with tech debt, it may take time
to identify the existing metadata you want the Agent to leverage. Remember, the Agent
is essentially reading your documentation to understand the capabilities of an action. If
that documentation is unclear, the Agent might be confused—or worse, disappointed!
So, some remedial work may be needed: better documentation, rigorous training, and
locking down those tricky edge cases.

Ultimate Guide to Creating Agents - Elements.cloud 14

 Channel: Determining Where the Agent Operates
Understanding where and how an AI Agent will operate is just as important as defining
its role. This requires a deep dive into your business processes to identify seamless
handoffs between AI and human agents. These processes will need to be updated
to reflect that an AI Agent is now part of the workflow. Whether it’s handling inquiries
through chat, voice, or other digital channels, the integration must be fluid. It’s not just
about automation; it’s about rethinking how you engage with customers to make every
interaction count.

Trust & Security: Permissions and Guardrails

Let’s face it—while AI Agents are smart, they still need guardrails. But you also need
security. This is where the Einstein Trust Layer comes in maintaining data privacy
and protecting you from toxicity. You’re giving these Agents the autonomy to make
decisions, so it’s critical to outline what they shouldn’t do. This goes beyond what you’d
explain to a new human employee, who would naturally err on the side of caution.
Additionally, you need to design a seamless handoff process for when the AI Agent
needs to pass a task to a human. We’ve all been frustrated by the experience of being
transferred from one human agent to another, only to repeat security questions and
re-explain the entire issue. This is your chance to create a flawless customer service
experience. How can you ensure the human agent sees all the prior conversations,
decisions, and data collected by the AI Agent? Your AI can even summarize it for them—
making the handoff smooth and efficient.

Ultimate Guide to Creating Agents - Elements.cloud 15

How Agentforce works
This is how an Agent works:
✓ T
 he Agent is launched from within Salesforce, Slack, an experience site, an external website, or
triggered from Apex or a Flow.

✓ The user engages with the Agent.

✓ T
 he Atlas Reasoning Engine interprets user input and reads the Agent and Topic information fields to
decide which Topic to use.

✓ O
 nce it has chosen the Topic, it looks at the Instructions and Action description fields associated with
the Topic to decide what to do each time based on the user input.

✓ T
 he Action is delivered by using the linked metadata which could be Apex, a Flow, a Prompt Template,
or an External Service.

Below is a Salesforce graphic showing how a single user question could go in a number of different
directions based on the data about Sarah’s missing order. The agent is deciding how to respond based
on the user input, the data and its instructions.

Ultimate Guide to Creating Agents - Elements.cloud 16

To deliver agents, Salesforce has created some new metadata types, and is reusing some existing ones.
The power of agents is the reusability of metadata, just like other areas of the Salesforce platform.
Architecting agents that are maintainable, secure and performant requires an understanding of the
metadata and dependencies.

Below is the structure of the metadata and the relationships between them. This is important to
know when you start planning your Agent. The diagram is an ERD, so you can see the 1:many and
many:many relationships.

Relationships of Agentforce, core Salesforce, and Data Cloud metadata

Let’s explore each of these new metadata types to understand how they are used.

Agents

Think of an Agent as a “domain”. A typical Agent is “Customer Support” or “Employee Support”. Initially,
Agents should be narrower in scope for the following reasons:

✓ Make experimentation easier

✓ Allow for test-led design

Ultimate Guide to Creating Agents - Elements.cloud 17

✓ Deployment faster

✓ Data governance and privacy simpler

✓ Monitoring more transparent

✓ Evaluation of ROI clearer

✓ Technical, business, regulatory, and reputational risks lower.

Currently, you can only have one Agent inside Salesforce (Einstein Copilot) for internal users. You
can have multiple Agents in Slack and on external sites, but don’t get carried away because you need
to simplify the user experience. As an example “Agent Astro” powers the help.salesforce.com, but a
different Agent is on salesforce.com. But it could be accessing the same underlying data.

For the future, think about the broad capabilities of the Agent e.g. Customer Support, but start with only
one Topic to keep the scope tight. Once you have deployed the Agent with the first Topic, you can add
more Topics to give the Agent more capabilities.

So in summary:

✓ Agent covers a broad scope: e.g. Customer Service.

✓ Agent supports a function or business unit.

✓ Agent has other information fields that determine its behavior. e.g. Tone of voice.

✓ Agent has one or many Topics.

Skills

Skills are not a metadata type. But you will hear Salesforce talk about Skills, which are a logical
collection of Topics. You could also think about Skills as a Use Case. For example: Financial Planner,
HR Recruitment Assistant, or FAQ/Policies.

Topics

Think of a Topic as the “Jobs To Be Done” that may currently be delivered by a human, that you want
to “agentify”. This may be by an Agent, or an Agent in partnership with a human. e.g. Answer a product
query, Arrange the return of a purchased product.

Here are some principles:

✓ Each Topic has a tight scope covering a “process with an outcome”.

✓ T
 he Topic Classification Description helps the Atlas Reasoning Engine decide when to use the Topic.

Ultimate Guide to Creating Agents - Elements.cloud 18

✓ T
 he Topic Scope, Instructions, and Actions help the Atlas Reasoning Engine decide how to deliver
the outcome.

✓ A
 Topic can be described in a process diagram. The diagram will help you to plan the activity
steps that will be delivered as Actions. It also helps identify the guardrails, and when the Agent
hands off back to a human. It also helps write any training or testing materials.

✓ A
 n ERD may be required for the Topic if there is data being pulled from different objects
or systems.

✓ Topics can have one or many Actions and Instructions.

Actions

An Action is how an Agent is able to interact with Salesforce to achieve its goal, based on its
understanding of the Action from its description.

Actions are reusable and are intended to be shared across Topics. Each Action is delivered through
a single Apex, Flow, Prompt Template, or External Service.

Here are some considerations:

✓ Actions should be narrowly scoped so that the Agent’s behavior is easy to predict.

✓ Ideally, reuse existing metadata, but ensure that the metadata only delivers the expected outcome
for the scope of the Action. You may need to refactor Apex and Flows, or build new ones.

✓ T
 he metadata descriptions are copied into the Action, but can be updated. These descriptions
inform the Atlas Reasoning Engine when and how to use the Action.

Instructions

The Instructions are stored with the Topic, rather than being a separate metadata type. They are rather
like prompts as they are written in natural language. Instructions are used to explain what to do and
what not to do. These provide information to the agent which it’s useful to think of as a series of “if X
happens (or is the state of what you find), do Y using Z”. In addition, guardrails provide explicit overriding
information of what not to do. Again, think of it as your bright, eager new employee. “Never do A under
any circumstances.” “If asked about pricing, never provide any details of any other customer deals”.

If each instruction is added as a separate item in Agent Builder, there is no control over the order that
the agent applies them. So if the order is important to get the agent to function effectively, you must
combine the instructions into a single instruction. If you want to be prescriptive about the outcome then
consider using an Action rather than an Instruction.

Ultimate Guide to Creating Agents - Elements.cloud 19

When writing Instructions:

✓ Instructions are part of the Topic, and not a separate metadata item.

✓ Instructions may reference Actions directly.

✓ Instructions are used to provide guardrails for the Agent, along with the Topic description and scope.

✓ The order of the Instructions changes the Agent behavior.

Metadata

Metadata is power behind the Salesforce platform and how Data Cloud and Agentforce are tightly
integrated. The article is a deep dive into metadata. This should be familiar to Trailblazers who are
becoming Agentblazers. Flow is so powerful there are fewer situations when Apex needs to be written,
which means that building agents is drag and drop. Metadata descriptions2 are important because they
are pulled through to the Action when the metadata is linked to an Action. You can edit them on the
Action so they are specific for the Topic.

Writing good descriptions is a key skill. AI can help you write those descriptions but you should be
selective. Do not run an AI app to document your entire org as you will need to validate every suggested
description. This will be a monumental task and much of the work is redundant. Instead, be selective.
Document just the metadata that matters at the time that you touch it.

2
Metadata documentation: the hero we don’t respect, but the one we need 20
Implementation
approach
Repeatable. Proven. Flexible.

We’ve developed an approach that has been documented as a UPN process map that is freely available
to download. You can modify it to dovetail with your implementation and governance processes.
Understanding the best way to build agents is still evolving, so the approach will be constantly updated
with attachments to reflect the latest thinking.

We’ve used the approach to build our first Agents. The first Agent took a few days because we were
setting up Agentforce, ensuring we had the right licenses, and documenting the approach. The majority
of the time was in design, ensuring that we were “building the right thing”. The actual build was very fast.
We’re now implementing additional topics and as we are building on our existing agent the speed
of implementation is remarkable.

Our experience of using the approach has created documentation templates, design patterns,
examples, and our internal Elements AI that accelerate the development. We used Elements.cloud as the
platform to design and build agents, so the documentation is version-controlled with access rights, and is
connected to the metadata. Now it is available for impact analysis as each subsequent use case builds
on the earlier implementations.

Implementation Approach: Top Level

Ultimate Guide to Creating Agents - Elements.cloud 21

Set up & Manage Agentforce

There are several licenses you need to run Agentforce, including Data Cloud. These are all covered in the
Trailhead and Agentforce workshops.

You need to set up an Elements Space and create a UPN process map for your Agent. We are calling this
map an AIM (Agent Interaction Map). The map has 3 levels:

✓ L
 evel 1 - List of Skills/Use Cases for Agent

✓ Level 2 - List of Jobs To Be Done (JTBD) for the Skills/Use Case

✓ L
 evel 3 - JTBD which is the overall process flow with instructions and actions

But the setup is not all technical. There is the natural inclination to rush into building Agents. Resist the
urge for three reasons.

✓ Y
 ou need to get buy-in from your sponsors for your Agentforce strategy. You will need to educate
them on the benefits, the implementation effort, and the pricing approach. Some organizations are
uncomfortable with consumption pricing because it makes budgeting unpredictable.

✓ Y
 ou need to assess your readiness; process, data, metadata

✓ Y
 ou need to understand and prove your internal Agentforce implementation approach. This is what
the free credits are for. Learning. Experimenting. Refining.

Ultimate Guide to Creating Agents - Elements.cloud 22

Identify Agent Skills / Use Cases & Jobs To Be Done

In this phase, you brainstorm the potential use cases based on their impact and potential benefits for
the business. You need to educate them on the “art of the possible”. At the same time, you need to
manage their expectations about the speed of delivery - especially the first user case. You need to select
the easiest and simplest for the first one to implement. You have a couple of approaches to identify the
use cases.

✓ Y
 ou can look at your existing processes, if they are documented, and work out which could be easily
delivered by an agent. It could be a customer-facing process that is knowledge-intensive; product
support. Alternatively, it could be an automated process like scheduling, canceling or rescheduling
bookings. Also, look at internal processes such as HR or onboarding. AI can help you document
your processes.

✓ L
 ook at processes that are delivered by your bots today that could be dramatically enhanced by
an agent.

✓ L
 ook in the JTBD library of diagrams that are pre-built.

✓ Look for inspiration from the Salesforce library.

An Agent will have several Skills / Use Cases. A Use Case will have several JTBD. For example, the agent
is Einstein Copilot which is the agent that runs inside Salesforce and is launched from the top bar. Within

Skills / Use Case JTBD

Vacation Management ✓ Book vacation

✓ Cancel vacation

✓ Check balance for you / team

✓ Check conflicts across team

FAQ / Policies ✓ HR Policy

✓ InfoSec Policy

✓ Acronyms/Terminology

Recruitment Management ✓ Register role

✓ Register candidate

✓ Schedule interview

✓ Issue offer

Ultimate Guide to Creating Agents - Elements.cloud 23

Create a UPN Process map for each agent. The Level 1 diagram has an activity box for each use case.
Each use case activity box drills down to a Level 2 diagram which has an activity box for each JTBD for
that use case. Each JTBD will be a Level 3 diagram and a JTBD will be delivered by a Topic within an
Agent. So the Level 3 diagram defines the Topic.

Design JTBD

In the rush to build an Agent, do not be tempted to skip

the design and try and build directly in Agent Builder. If
you do not understand the process you will spend days “We have been promoting
of frustration trying to get the agent to work consistently.
the whole notion of design
With an Agent you are essentially programming an
app using natural language. And language can be very
thinking and process thinking.
ambiguous. The build may be fast without time spent When it comes to AI, having
designing the process but the testing will be never ending. the mindset of process
Worse, you may never be confident that you can deploy
pattern thinking, I think it’s
the agent into production.
going to be really critical,
From experience of building our own Agents, we’ve especially as we are held
realized that you need to think through the process in accountable to having ROI
far more detail than you would if a human was delivering
at the end of the day with
the process. The AI agent doesn’t have the company /
contextual common sense. It only knows what you’ve told
improving these technologies
it about. You need to be far more explicit about the rules to improve processes.”
and guardrails.
Juan Perez
A process diagram is a far easier, more visual way of CIO of Salesforce
understanding the scope of the JTBD. You can then see
where you need Actions. You can identify the hand-offs
back to a human.

Now that Element’s AI can generate the agent instructions, guardrails and test utterances from the UPN
diagram, this is the most important phase. The design phase can be short, but make sure that you’ve
considered all the angles.

Our first Agent was allowing employees to query HR Policies based on their location, which was
straightforward. Our second was more complex: booking PTO for an employee. We have different policies
based on the employee’s location in the world. And there are rules about planning PTO; you can’t exceed
your allowance, you shouldn’t book weekends or their local country public holidays etc. The process
identified that to make the agent work we needed to update some metadata, create some new metadata,
and restructure some of our policy documents. None of this was impossible, but it was far easier to get
right before we started building the Agent as it affected how we wrote the instructions.

Ultimate Guide to Creating Agents - Elements.cloud 24

 Level 3 UPN diagram, drawn using Elements.cloud

If you operate in a regulated industry, the UPN process diagram in Elements can provide governed
documentation of the processes the Agent is supporting, with version controlled content and approval
histories to satisfy compliance requirements. The diagram and their notes can be signed-off and
version-controlled, so that you build up a history of the changes. This will prove vital as the agent will
iterate over time.

There are 4 ways to create that JTBD Level 3 diagram. We and consultants are building template
diagrams for different JTBDs. These can be imported and used as the basis of your JTBD diagram.
Elements AI will draw a diagram from your org metadata. You could use Elements AI to draw the diagram
from a sketch, image, diagram, script or SOP. Or you can start with a blank sheet in a live workshop
with your stakeholders. Whichever approach you use, it is the quickest way of designing your agent and
getting consensus.

As we all learn how to make Agents deliver great results consistently, the Instructions and Actions
descriptions will keep evolving. Using the diagram snapshots and versioning enables you to track what
works, and what didn’t in the past. This makes troubleshooting and testing so much easier.

Elements AI can run a consistency check to ensure that you have followed the standards, and not
introduced conflicts, duplications or gaps. It can also suggest where you can streamline the process.

Ultimate Guide to Creating Agents - Elements.cloud 25

For the Agent to deliver the Actions, you may also need to build or modify existing metadata - Apex,
Flow, Prompt Templates. Elements AI can write user stories for this work, along with acceptance criteria
for testing. It can also suggest metadata to reuse or what new metadata is required. If you are able to
reuse metadata, then make sure that it is documented and you understand exactly what it does and you
are happy to let the agent loose using it.

Again this all sounds time-consuming. It is not. The UPN diagram is how you are visually designing and
building the Agent. Mapping the process can happen very quickly, and the UPN diagram is a great way of
accelerating executive buy-in, and sign-off from your security and compliance teams.

Build Topic

Whilst you are deploying an agent, you are actually building a Topic that is part of the agent. The Level 3
UPN diagram of the JTBD is the end to end process definition of that Topic.

To build an agent, first you create the agent in Agent Builder. Skip this step if the topic is being added to
the Einstein Copilot, or you are adding a new Topic to an existing Agent.

Inside the Agent you need to create the new Topic and add the scope and classification descriptions.
Elements AI can write the Instructions and Guardrails and save them in the clipboard so that you can paste
them straight into Agent Builder for the Topic. Elements AI also takes a snapshot of the Level diagram so
you have a change log.

We have discovered that to get consistent and accurate results from the Agent you add only ONE
instruction that contains all the instructions and guardrails created from the Level 3 diagram. Discovering
this has transformed our ability to build agents that work as we expected, and that has dramatically
reduced the testing time. Also the better the analysis and UPN diagram in the Design phase, the better
the Agent.

You need to build any new metadata (Flow, Apex, Prompt Builder) before you can add Actions to the Topic.
The user stories created in the previous phase define the metadata you need to build or existing metadata
you need to modify. Once the metadata has been created you can add Actions to the Topic and the Actions
reference the metadata. The Action descriptions are copied from the metadata descriptions, but can be
fine-tuned for the Action.

Test Topic

Firstly, the Elements AI generates test utterances for each process path on the Level 3 diagram. It is
very good at creating utterances that consider different perspectives, many of which we would not have
thought of. This is a huge time saver.

Ultimate Guide to Creating Agents - Elements.cloud 26

 The power of this approach is:
These utterances are complementary to test
center, where they can be run and evaluated
automatically. Test Center will create general Speed: Instructions and tests
generated for Agent Builder.
utterances, but Elements AI considers the
process perspective.

Because building an Agent is so fast you can

Reliability: The Agent reflects
iterate quickly. As you test the user input and
the process.
see how the agent responds, you will make
changes. These changes are made to the
Level 3 diagram and then you regenerate the
Governance: The UPN diagram is
instructions to copy into Agent Builder. You also
the version controlled back-up of
update the test utterances. And the diagram the Agent design.
automatically creates a change log.

Deploy Agent

Deploying the Agent is no different than pushing a new app into production. You need to deploy the
metadata, migrate data, and communicate the changes. At the same time you version the Level 3
process diagram.

This is a critical part of the governance of an agent. Arguably an agent needs more oversight and
governance than humans do because they have no organizational context. But on a positive note, the agent
will always be referencing the most up to date corporate policies and processes. So it will be more diligent.
Make a change and instantly your agent is following it.

The Agent and its related metadata - Topic and Actions - and any other metadata are migrated from the
Sandbox to Production just like any other Salesforce metadata. At the same time you need to ensure the
data you need is migrated to production. For example, if your JTBD is Accessing product support, then are
your support articles in the knowledge base in a format that AI can read?

When the Agent is built in the Sandbox and deployed to Production, the metadata is visible in the Elements.
cloud metadata dictionary and all the dependencies are tracked. The metadata can be linked back to the
Level 3 diagram action activities and the User Stories. With the reuse of metadata, it is critically important
that you know the impact of changing something.

Ultimate Guide to Creating Agents - Elements.cloud 27

 Dependency Tree in Elements.cloud metadata dictionary

“For all of those of you who have been For example, you want to change a Flow
fastidiously documenting…. You are that supports an internal process. That
same Flow is also used in 3 Topics,
done. And for the slackers out there,
as you can see from the image above.
you have another reason to add That change could impact the agent’s
descriptions. You give a description behavior. Or worse, the agent will ignore
to an action, just like you would it and come up with a workaround. And
how would you even know the agent
introduce it to a colleague. That
was doing it, because the change won’t
teaches Einstein. Let’s give it up for “break the agent”?
documentation. I love it.”
Repeat after me: dependencies and
John Kucera, documentation3.
SVP Product Management,
Salesforce

3
Metadata documentation: the hero we don’t respect, but the one we need 28
Operate & Monitor

The impact of consumption pricing is a source of debate. It is a fundamental shift. But less has been
thought about monitoring. This is no different from supervising a team of call center agents. Except you
have far more granular data on performance. Every conversation between the Agent and a user is logged
and what Action was used for each step is stored in Data Cloud.

Monitoring the effectiveness of the

Agent and the handoff points back to
a human will help scope the additional
capabilities to be added to Topics, or
new Topics that need to be added to
the Agent. It also means you see where
to train and update the Agent.

Armed with this data you can evaluate

the ROI against your success criteria.
That means you can understand your
monthly bill, but then forecast future
usage and savings. That will give your
executive leadership the confidence to
double down on AI agents.

Ultimate Guide to Creating Agents - Elements.cloud 29

Success secrets
Go slow to go fast

Agentforce will deliver results faster if you have a mature implementation approach; process-led change,
data governance, and metadata management. Use the first Agent to put these in place to build a strong
foundation. This will accelerate subsequent Agent use cases. It feels like you will slow you down, but it is
actually faster to get adoption4:

It starts with well-defined processes: Agentforce is a process-led implementation. Organizations

that already documented processes can quickly decide what to “agentify” to provide an
exceptional user experience.

Strong data governance is key: Bad data is

junk food for AI. Control over data quality “We took longer to build our
and governance will be a differentiator and
first agent because we took
impact the speed at which agents can be
deployed and the accuracy of the support time to get the approach right.
they can provide. The second, more complex
agent took less than 5 hours.”
Metadata is how agents know your
business: Agents use metadata to drive Adrian King,
actionable outcomes. And they use Founder and CTO
metadata descriptions to understand Elements.cloud
what actions to take and when.
Metadata documentation used to be
an afterthought. That has to change.

The organizations that have these disciplines mastered will be able to deliver agents more quickly. They
will accelerate ahead of their competitors. So, even if your organization does not feel ready to implement
an Agent, then now is the time to start working on these best practices.

It all starts with process. Shift Left.

The initial agents are replacing or supporting business processes that are currently delivered by humans.
Those processes need to be clearly defined down to a level of detail that means you understand how the
Agent can be built. The process diagram is the Topic that will identify the Actions and how they are likely
to be delivered. The process diagram will help write the Instructions by showing the key steps and the
handoffs back to a human. The devil is in the details.

4
Yes, just like driving behind a cautious driver when you’re late—turns out, they might be onto something. 30
Don’t think that the process mapping will slow down the delivery of an Agent. The process diagram
can be created very quickly. The first draft could be drawn from AI by looking at your org, existing
diagrams, or even the text from a transcript.

If you don’t draw the process diagram, you will take far longer to deploy an Agent due to the constant
rework as you discover things you have overlooked. Testing will take forever as it is difficult to
understand which instructions are causing the problems. Building an agent using the Agent Builder is
really really fast, once you know what you are doing. If you build the wrong thing your brand will suffer,
you won’t get the ROI you expect, and you will drop behind in competitiveness.

Take the time to plan.

Start small. Think big.5

Building agents will be iterative. The initial Agent is deployed with just one Topic. You can then start
extending the capabilities by adding Actions to Topics, and Topics to Agents. You need a repeatable
approach that provides enough analysis and documentation to ensure that the Agent delivers an
exceptional user experience, but is lightweight enough that it does not slow the pace of innovation.

Focus on narrowly scoped use cases to build experience. This could be customer-facing support sites,
or employee-facing business processes. The learning you get from this first Agent is as important as
the ROI.

Avoid Data Cloud complexity

Data Cloud is required for every Agentforce implementation because that is where the tracking data
is stored. You don’t need to implement Data Cloud unless you want to pull data from external systems
and make it available to the agent via Apex, Flow or Prompt Templates.

If your agent is using unstructured data - support articles, policy documents, contracts - then it will
need to use the Data Cloud’s vector database and Prompt Templates to query it. This is a straightforward
use case.

More complex Data Cloud use cases will slow down the implementation of your agent. This will rob you
of valuable momentum and learning. What complicates any implementation is harmonizing data from
multiple external systems, for example: getting a single, aggregated view of a customer who is using
different login identities across your customer-facing systems - marketing, sales, order, fulfillment,
support, and finance.

5
Kind of like how a small seed becomes a massive oak tree—unless it’s one of those Bonsai trees 31
Agents can help build agents

Elements.cloud provides the design documentation and metadata insights that are the foundation for
exceptional agents. We’ve designed a repeatable implementation cycle with documentation templates.
This standardization has enabled us to build AI agents that can accelerate the implementation cycle.

AI agents can:

“Using a process-led ✓ Draw process maps.

approach we went from zero ✓ Highlight areas that could be agentified.

to a working agent to support ✓ Write instructions and create actions.

staff around HR Policies ✓ Check on conflicts between instructions,
in less than 2 weeks from actions and your guardrails.

Agentforce going GA.” ✓ Write user stories and acceptance

criteria.
Jack Lavous,
✓ Identify metadata that could be reused.
VP Business Excellence and
Operations, Elements.cloud ✓ Write instructions.

✓ Create training and testing content.

AI can dramatically reduces the time to build an Agent. But your knowledge of your business, your needs,
and your expected outcomes will be required to update the results.

Ultimate Guide to Creating Agents - Elements.cloud 32

Using Elements.cloud
The approach can exploit the process-led change
(diagramming, user stories) and metadata management
(dictionaries, dependencies) capabilities of the platform.
Having all the information in one place enables you to apply
AI to accelerate the build of Agents. But it also provides
version-controlled documentation that enables you to have
confidence as your Agents evolve rapidly. Using Elements
you can also take advantage of the library of prebuilt JTBD
Templates which are UPN diagrams with Instructions and
Actions that will be developed by us and consultants.

Finally, it provides the support you need to put the

foundations in place that stretch beyond the Agentforce
implementation; well-defined business processes, strong
data governance, and comprehensive application metadata
management. So even if you are not ready to implement
your first Agent, you can still use Elements to help you get
ready. Mastering these disciplines, if they are alien to your
organization, cannot be achieved overnight.

The alternative is diagrams and spreadsheets that will

rapidly get out of control once you have delivered the first
Agent with the first Topic. The implementation approach can
be applied using the Elements Professional license type. A
small price to pay for the control, confidence and peace of
mind it gives you.

If you are only planning to build one Agent with one

Topic, then it is clearly not necessary. But the potential of
Agentforce is a small army of Agents across your business
delivering all the boring, tedious, repetitive work - quickly,
reliably, 24x7.

Ultimate Guide to Creating Agents - Elements.cloud 33

Final word
Agents are as disruptive to the delivery of apps and their underlying business models as the shift from
on-premise to cloud was over 20 years ago.

The early cloud use cases were limited but drove benefits. However, they did not give us any insights
into what was going to be possible. It feels like the agents are following a similar path, except it is all
happening so much faster.

Those organizations that are best placed are already in good shape; well understood processes, strong
data governance, and effective metadata management. But they need to lean-in and explore the art of
the possible.

No organization can afford to use compliance, data quality, or risk as the excuse to sit on the sidelines.
The future winners have already started; they are piloting, experimenting and learning. And they are
accelerating away from the pack who may never catch up.

Elements.cloud is the Change Intelligence

platform for Salesforce. 


We help organizations understand their business and systems configuration and then ability to
apply that intelligence to decide how best to implement change.

As part of our commitment to the Salesforce community, we are publishing regular reports
on trends in Salesforce Orgs across key areas of complexity, documentation, technical debt,
adoption and overall best practices in Org Management.

Talk to us Learn more

Ultimate Guide to Creating Agents - Elements.cloud 34