Automation · Nov 4, 2026
Before AI agents can run your business, your business has to agree on its own words
Ask three people in your company what counts as an "active customer" and you may get three answers. Sales counts anyone with an open deal. Accounting counts anyone who paid in the last 90 days. Support counts anyone who called this year.
People work around that every day without noticing. An AI agent cannot. It will pick one definition, act on it confidently, and hand you a report or send an email based on a number nobody actually agreed to.
That problem was the quiet headline of this week's AI announcements.
What shipped this week
On November 3, at its Expedition 2026 event, Snowflake announced always-on agents that work in the background and check in only when a decision or sign-off is required. Alongside them it announced Business Ontology, which it describes as one governed place to record what business terms officially mean, serving only approved definitions to agents. It also announced Cortex AI Gateway to govern how agents access models, call tools, and track spending.
The same day, monday.com used its Elevate '26 event to describe its AI workspace around governance. In its words, you choose what AI can access, down to a single board or Slack channel, and nothing is accessible by default. It also added per-user spending limits and alerts before a budget runs out.
SAP made Cohere's agent platform, North, generally available on SAP Business AI Platform, aimed at regulated industries that need agents and workflow automation to run inside their own infrastructure. And on November 2, Abacum launched Abra, a planning agent where, as the company puts it, proposals wait for approval and every step is on the record.
Four companies, four different markets, one pattern. The agent is no longer the hard part. The context and the controls around it are.
Why definitions come before agents
Centervert grew out of exactly this kind of problem. Before it was a company, the work that started it was pulling business data scattered across two project management tools into one CRM and ERP, so there was a single place to answer basic questions about customers and work. That consolidation mattered more than any single feature.
AI agents make that kind of cleanup more valuable, not less. An agent that drafts invoices, answers customer questions, or flags overdue accounts is only as good as the data and definitions underneath it. If "overdue" means 30 days in one system and 45 in another, the agent will be wrong in a way that looks right.
OpenAI made a related point this week from a very different field. Describing a test device its GPT-6 Astra model helped design, the company wrote that getting it ready for manufacturing still took several rounds of review, because the model did not consistently account for the manufacturer's production limits and tolerances. Capable models still miss the rules nobody wrote down.
You do not need an enterprise data platform to do this
The tools announced this week are built for large companies. The underlying discipline is not. A small or mid-sized business can get most of the benefit with a few practical steps.
Write down your ten most important terms. Customer, active customer, lead, job, open ticket, overdue, margin. One sentence each, agreed by the people who use them. A shared document is fine to start.
Pick one system as the source of truth for each. Customer records live in the CRM. Invoices live in accounting. When two systems disagree, everyone, including any agent, knows which one wins.
Give agents the least access that works. The monday.com default of nothing accessible until you allow it is a good rule anywhere. An agent that summarizes support tickets does not need your payroll folder.
Decide where a person signs off. Reading and drafting can often run on their own. Sending money, emailing customers, or changing records should wait for a human until the agent has earned more trust.
Set a spending limit and watch it. Usage-based AI pricing is now common. A budget alert is cheap insurance against an automation that loops.
Where Centervert fits
This is the work we like best, because it is where technology starts working together instead of piling up. We help businesses map their processes, clean up and connect the systems they already run, and set up automations on top, using tools like n8n, Zapier and Composio where they fit. Our Managed IT side keeps the access controls, accounts and security underneath it all in order.
Our house rule is Think First, AI Second. Before we put an agent on a process, we make sure the business agrees on what the process actually is.
The one thing to do this week
Get the people who run sales, operations and accounting in a room for 30 minutes and agree on what "active customer" means. Write it down. If that conversation is harder than you expected, you have found the first thing to fix before any AI agent touches your business. If you would like help working through it, we would be glad to talk.
Sources
- Snowflake CoCo and CoWork Power the Agentic Enterprise, Snowflake
- Snowflake Helps Enterprises Deploy AI Agents with Trusted Business Context and Built-In Controls, Snowflake
- Updates from Elevate '26, monday.com Blog
- SAP and Cohere Empower Regulated Industries to Innovate and Deploy AI Faster, SAP News Center
- Abacum Launches Abra, Abacum
- Accelerating engineering for the physical world, OpenAI
