Most teams have rolled out an enterprise AI tool and given everyone the mandate to figure out how to use it. There might be a monthly show-and-tell or some departmental enablement sessions to keep things moving.
This makes individuals more effective. But it doesn't change what the company is capable of.
The teams pulling ahead are doing a few things differently. Here they are:
1. Publishing data sets
AI needs context to operate. That context is your company's data. CRM records, marketing campaign history, employee rosters, product usage, compliance data, billing records, call transcripts, playbooks... the list is long.
Most of it falls into two buckets. Transactional data that captures what the business has done over time, and operational knowledge about how the business works. Both need to be identified, validated, and published in formats that agents can actually read.
2. Documenting the company ("as code")
Operational knowledge needs to be consolidated and made available to the people and agents working with transactional data. Design templates, delivery standards, process frameworks, brand guidelines, etc.
I call this "company as code." (Didn't come up with this, just commandeered it.)
We always tried to do this with documented procedures and internal wikis. Great instinct, but limited by human capacity to follow and maintain those things over time. We can now use learning loops to address those limitations (see #6 below).
3. Demanding AI-ready systems of record
A number of pillar systems of record are worth continued investment. CRM, ERP, marketing automation, HRIS, etc. But the best vendors in these categories now offer modern APIs, MCP and CLIs (command line interfaces), and clean data export options.
Keep in mind, all of these platforms are shipping their own embedded agents. Fine for work that stays inside a single system. But be selective about opting into those capabilities. Most of the real work agents can do will span departments and systems. Don't get stuck in an ecosystem that doesn't serve you well.
4. Giving employees the ability to build agents and apps
Using centralized data to generate one-off outputs that get shared in Slack or dropped into Notion, SharePoint or GitHub is a great start. But the next level is employees building their own agents and applications on top of shared data and systems the org has published.
Letting individuals build in a safe, sanctioned way on shared data sets is an incredible way to unleash creativity and knowledge that's locked up in your people's heads today. Everyone can build in this model, and the best agents and apps become candidates to roll out to the whole team or company.
5. Letting builders deploy safely
Agents aren't always the answer. When a process is predictable, it's smarter to have an agent build an app once and deploy it. Tokens burn every time an agent runs, and a deployed app usually runs on far fewer, or none at all.
Give power users the sandboxes and platforms they need to ship apps safely and securely.
6. Building continuous learning loops
Few companies are doing this yet, but it's where things are headed.
Every time we record a call with a customer, several things need to happen. Update the CRM record. Log contact activity. Capture customer sentiment. Update competitive battle cards. Track product requests. Flag issues to the right people. Draft a follow-up to the customer.
Agents can take action, but the company continuously refines its skills and processes so that it improves with every customer interaction.
This is what "multi-player AI" looks like in practice. Where does your team stand?
- Are you doing any of this today?
- Which layer are you getting stuck on?
We're helping companies work through each of these layers. If you want to talk through where you are, we're here.