That gap is the whole problem, and it is not solved by better prompts or another license. It closes when AI has your company’s context, sits inside a process somebody owns, and produces work that goes somewhere. We build that, with Anthropic Claude for the reasoning and ServiceNow for the process and the record.
Pulse & Patterns is a registered partner in the Claude Partner Network.
We build Claude-enabled solutions at the intersection of Anthropic Claude and ServiceNow.
Not sure which AI work is worth doing yet? Start with AI Use Case Discovery. Looking for AI inside your ServiceNow processes specifically? That is ServiceNow AI.
Most organizations are eighteen months into AI adoption and cannot name a business process that runs differently because of it. The licenses are bought, the enthusiasm is real, and the operating result is nowhere.
It writes well and it has no idea what your product is called internally, which of your entities owns which customer, or what your approval threshold is. So every useful answer needs a human to supply the context first.
Someone generates something good at a chat window, then copies it into the system where the work actually lives. The AI saved ten minutes of writing and added five minutes of transcription.
Fifty people have each worked out their own way to summarize a document. None of it is shared, reviewed or reusable, so the organization learns nothing and the quality varies by person.
AI adoption is measured in seats and enthusiasm because no business owner ever agreed which number was supposed to move.
Decide what it is for
Give it the knowledge
Put the output where the work lives
Agree what a person checks
We build AI into a specific business process, with your organization’s context behind it and a named owner in front of it. That means deciding what the AI is for, giving it the knowledge it needs, putting its output where the work already lives, and agreeing what a person still checks.
The technology matters less than that sequence. You should not need to know what a skill, an agent or a protocol is in order to understand what you are buying. You should need to know which process gets better, who owns it, and what a person still approves.
So we do not sell Claude features. We build a named outcome on a named process, and we show you the same capability running in our own operations first.
Every frontier model is capable enough for almost every business task an organization actually wants done. Almost none of them work well inside a company, and the reason is always the same: the model has no idea how this particular business operates.
None of that is in the model, and none of it arrives by writing a longer prompt.
01 Your terminology
02 Which system holds the master version of a record
03 Who signs off what
04 How your entities differ
05 What last quarter’s decision actually was, and why
Making that context available, curated and current is its own piece of work: Company Second Brain.
We answer this with what is running, not with adjectives. Ask any partner the same question, and if you get a capability list back instead of something they can show you, keep asking.
We run Claude inside our own company across knowledge, documentation, delivery support, profiles and presentation work, on a shared context layer with reusable skills and playbooks rather than fifty private habits. This is the capability we can show you working, today, because it is ours.
Knowledge migration and curation, requirements and documentation work, and AI-assisted configuration against a development instance with approvals and a build log. This runs in live client delivery, and we will walk you through it.
We build agents that are permissioned, logged, reversible and stoppable, and we run them: in our own operations and in the ServiceNow environments we build in. The boundary is written down and tested before an agent goes near a process that matters, which is the part most teams discover too late.
Which piece of work is slowest in your business?
We will tell you honestly whether AI is the answer.
Five steps, and the first two are where the value is decided.
Which process, which measure, whose job it is.
If nobody will own the result, we stop here and say so.
What the AI needs to know, where that knowledge lives today, whether it is good enough, and who keeps it current.
This is usually the largest piece of work and it is almost never what a client expects to be buying.
What the AI drafts, what it decides, what it may change, what a person approves, and what gets logged.
Written down before anything is built.
Output lands in the system of record, not in a chat window. Usually that means ServiceNow, sometimes it means somewhere else.
Occasionally it means we tell you the process is not ready.
One team’s solution becomes something the next team can use, with the standard, the review and the ownership attached.
Otherwise you have bought a clever thing that one person maintains.
Our own company context sits in a structured layer that Claude can use: how we describe what we do, our terminology, our operating rules, our templates and our design standards. Our people work from shared, reviewed, reusable assets rather than each inventing their own approach. Our delivery documentation, requirements work and knowledge migration all run through it.
This is where our specificity comes from.
We know which parts work first time, which take three attempts and which are not worth building at all, because we built them on our own business before bringing them to yours.
When an organization asks us how to get from scattered individual AI use to something the business owns, we are describing a route we have walked rather than a framework we drew.
RELATED Company Second Brain on Claude
Neither half is much use to a business on its own.
Reads
Reasons
Drafts
Analyzes
Holds no record of your business and no authority to act in it.
Process
Records
Permissions
Approvals
Audit trail
Does not reason.
01 Understood
02 Checked
03 Routed
04 Actioned
05 Recorded
Put them together properly and a request can be understood, checked against what a customer is actually entitled to, routed, actioned and recorded, with a person approving whatever needs approving. That is the architecture worth building.
ServiceNow documents Claude as a model provider and as a client of the standard protocol that lets an AI assistant use approved business systems. It is not an agent platform that ServiceNow discovers on its own. And what any given organization can connect depends on its ServiceNow release, its edition and its configuration, so we establish that against your instance rather than answering from an architecture diagram.
Claude is our preferred model and the architecture we build does not force every use case onto one model, because the model that reads documents well is rarely the one you want holding write permissions.
RELATED ServiceNow AI AI Control Tower ServiceNow Integrations
An agent being the wrong thing to run in a process almost never means AI has no place in the work. Usually it belongs in how the solution gets built rather than in what runs every day, and we use Claude heavily on our side of that line.
If the trigger, the logic and the output are all known, deterministic automation is cheaper, easier to govern and does not vary.
We still build it with Claude: reading the existing rules, drafting the configuration, generating the test cases.
Two systems that do not talk to each other need an interface, not an agent.
Claude accelerates that build, drafting and testing the API work or an MCP server against your instance, and what runs in production is a clean integration you can support.
AI on a process nobody has redesigned makes the wrong thing happen faster and harder to unpick. Redesign comes first.
AI earns its place there: process discovery across your real execution data, variant analysis, and a redesign your team can review instead of wait for.
If nobody can say which system holds the truth for the records involved, that is the first project.
It is also a project Claude is genuinely good at: extracting, structuring, classifying and de-duplicating what you already have, so the foundation gets built rather than deferred.
Some work is valuable precisely because a person is accountable for it. Keep it human, deliberately.
Use AI to prepare the decision rather than to make it.
Discovery decides what is worth building and what ready means. Implementation builds the agreed capability.
This page answers what intelligence and knowledge-work capability we bring. ServiceNow AI answers how that intelligence gets enterprise context, permission and a system of action.
AI Use Case Discovery decides which AI opportunities are worth doing. Anthropic builds the ones that need reasoning over your own knowledge.
The Anthropic hub builds AI into your business process. Claude Enablement builds your people’s ability to use it.
Our own operations use Claude daily, on shared context and reusable assets. That is where our opinions come from, and it is why we can tell you which parts are harder than they look.
Most AI consultancies do not know your ServiceNow platform, and most platform partners treat AI as a feature. The value sits exactly where those two meet, and that is the whole reason this practice exists.
No account layer, no handover to a team you have never met. The people who design it are the people who build it.
Tell us which piece of work is slowest and we will tell you honestly whether AI is the answer, whether it is automation, or whether something has to be fixed first.
Three things that are rarely about the model. The AI needs access to your organization’s actual context: terminology, processes, records, decisions. Its output has to land in the system where the work lives, not in a chat window. And a named business owner has to agree which measure should move. Organizations that skip the context work get impressive demonstrations and no operating change.
Using it is individual: a person, a conversation, a better draft. Implementing it means the capability belongs to the organization rather than the individual. Shared context, reusable assets, output in the system of record, a defined boundary between what the AI does and what a person approves, and an owner accountable for the result. The first improves someone’s day. The second changes how a process runs.
Claude provides reasoning over language and knowledge. ServiceNow provides the process, the records, the permissions, the approvals and the audit trail. ServiceNow documents Claude as a model provider and as a client of the standard protocol for letting an AI assistant use approved business systems. What can actually be connected depends on the ServiceNow release, the edition and the configuration, so it is established per instance rather than assumed.
Use ordinary automation when the trigger, the rules and the output are all known and stable: it is cheaper, faster, easier to govern and it does not vary. Choosing the right kind of AI for each use case is part of AI Use Case Discovery.
Not all of it, and that expectation stops more useful work than it protects. How we check whether a process and its data are ready is part of AI Use Case Discovery.
That depends on your configuration, your obligations and your controls rather than on a general answer. The questions worth asking are what the model may retain, where processing happens, which systems it can reach and acting as whom, what gets logged, and who can stop it. We design those controls before anything goes into a real process, and we establish the specifics against your requirements rather than quoting a datasheet.
By treating the useful ones as company assets rather than personal habits. A shared context layer so everyone works from the same version of the truth, reusable assets with an owner and a review, and a standard for what good output looks like. Otherwise the quality varies by individual, nothing improves cumulatively, and the capability leaves when they do.
Technically, in narrow and well-bounded cases. Whether they should is a different question, and it turns on permissions, reversibility, auditability and who is accountable when it goes wrong. Our position is that agency raises the weight of every control that already existed rather than introducing a new category. We would start with a bounded piece of work, a person approving what matters, and a tested way to stop it.
The AI model itself, for example Claude.
The one system that holds the master version of a piece of data.
One piece of AI given a job, a set of tools and a boundary.
Your own copy of the platform.
