- Towards AGI
- Posts
- Databricks Just Gave Its AI Agents A Night Shift. Is Your Data Ready?
Databricks Just Gave Its AI Agents A Night Shift. Is Your Data Ready?
what always-on agents will expect from your enterprise data
Today, we're diving into:
Gen AI: Databricks launched an always-on Genie agent that works in the background and reads your company through a new Ontology feature
Hot Tea: CoreWeave is asking AI customers to pay upfront, with about $9 billion in prepayments and deposits expected over the next three months
OpenAI: OpenAI bought Odessia, a year-old AI travel agent startup, and its founder is now working on the future of travel
Databricks Just Gave Its AI Agents A Night Shift. Is Your Data Ready?
Databricks has launched agentic tools that continuously analyze and organize the data scattered across a business, Fortune reported on November 11. The updated Genie assistant runs in the background, shadows an office worker through the day and makes suggestions before it is asked.
A separate feature called Ontology scans data across the organization, from reporting lines to department projects, so one person can pull information or run an analysis that used to need several teams.
Your Agents Will Only Be As Smart As Your Data.
Find out whether your data can support an agent that never switches off.
What Databricks Actually Launched
The new Genie can take on multistep jobs that need extended work to finish, TechCrunch reported. It works through an ontology Databricks built to give the agent context from the data customers already store on its platform. Leaders can assign it tasks and, according to the company, it completes them unsupervised.
"The agents are running in the background all the time. So it's always on." Ali Ghodsi, CEO, Databricks
Ghodsi told Fortune he uses it every hour of the day. TechCrunch watched him run it on an employee satisfaction survey, a failed sales pitch and podcast preparation.
The launch landed a day after Anthropic released its own office productivity tools. The race for the knowledge worker's desk is now a race to sit inside your data.
Who This Is Built For
Ghodsi pointed to Dunbar's number, the idea that people can keep track of about 150 relationships. His argument is that once a company passes that size, nobody can hold the whole picture in their head, so an agent that can read the organization becomes useful.
Databricks itself has about 12,000 employees. If its own CEO needs an agent to see across the company, so will yours.
This Is Not An Agent Problem. It Is A Context Problem.
The easy reaction is to treat this as another chatbot upgrade. That misses what changed.
An assistant you query waits for a question, so a person checks every answer. An agent that runs all day acts on whatever your data says, including the stale definitions, the duplicate customer records and the dashboards two departments calculate differently.
The real bottleneck is not the model. It is whether the context the agent reads is trusted, owned and current.
An ontology is only a map. If the map is wrong, an always-on agent follows it faster and more often than any person would.
The wrong reaction is to switch on background agents for everyone and fix data issues as they surface. By then, the agent's suggestions will already be in meetings and decks.
The better alternative is to decide which data an always-on agent may treat as the truth before it starts working the night shift.
What You Should Do This Week
Name your source of truth. For your ten most-used metrics, decide which system owns the definition.
Map who owns what. Every dataset an agent reads should have a named owner who can fix it.
Start with read-only. Let background agents suggest and retrieve before they act.
Log what it used. Every suggestion should show which data it came from, so people can check it.
Building that context starts with trusted enterprise data. DataManagement.AI helps organizations create that foundation by unifying metadata, lineage, governance, business glossaries and knowledge assets into a single intelligence layer. It automatically discovers metadata, maps lineage across the modern data stack and identifies downstream dependencies, creating a centralized knowledge layer that stays continuously updated as your environment evolves.
The organizations moving first will not just turn on always-on agents. They will make sure every agent reads from data the business has already agreed is true.
CoreWeave Wants Your AI Money Upfront. Is Your Budget Ready?
CoreWeave is asking customers to pay in advance, borrowing against specific contracts and selling more short-term capacity at higher prices, TechCrunch reported on November 11 after the company's third-quarter earnings call.
Revenue rose 168% year over year to $3.66 billion. The company still posted a net loss of $709 million, with $846 million in quarterly interest expense.
What Changed In The Deal Terms
Prepayments: about $9 billion in prepayments and deposits expected over the next three months, including more than $3 billion tied to new contracts for capacity already built.
Price rises: about 25% in July, then another 10% later in the quarter.
Short-term premium: short-term capacity sold at more than $40 million per megawatt a year, against $20 million to $25 million for four- and five-year contracts.
CEO Michael Intrator said prices are still climbing this quarter and "more than offset" cost increases. He also said CoreWeave will not use the unsecured debt market at current yields.
Why This Lands On Your Desk
You may never sign a CoreWeave contract. Your model providers and AI vendors probably rent from someone who does.
This fundamentally changes how you should read your AI costs. When infrastructure providers ask for cash upfront and charge a premium for flexibility, those terms flow downstream into what you pay for inference.
The Wrong Reaction Is Locking In Everything Now
Signing long, prepaid commitments to beat the next price rise feels prudent. It also ties your budget to today's models and today's workloads, at a time when both change every few months.
The better alternative is to know exactly which workloads need guaranteed capacity, and which can run on cheaper or smaller models when prices move.
Know Your Inference (KYI) is built for exactly this decision. When you measure how accurate, fast and costly each AI workload is on your own cases, you can commit capacity where it earns its keep and keep everything else flexible.
Ask your vendors where they run. Know which compute providers sit behind your AI contracts.
Read the price-change clauses. Find out whether your rates can move mid-term.
Split steady and spiky work. Commit only for workloads that run every day.
The organizations moving first will not just secure more compute. They will know which work deserves it, and keep the rest free to move when prices do.
OpenAI Just Bought An AI Travel Agent. Who Books Your Next Trip?
OpenAI has acquired Odessia, an AI travel agent startup, TechCrunch reported on November 10. Founder Francis Davidson, who previously founded Sonder, said he is now working at OpenAI on the future of travel.
Terms were not disclosed, and TechCrunch described it as looking like an acqui-hire. Odessia had raised a $6 million pre-seed round led by Sequoia and launched its agent about five months before the deal. The product has already been shut down.
Why A Small Deal Matters
"We're thrilled to welcome Francis to OpenAI." OpenAI spokesperson
OpenAI has not said what it plans to build. But the signal is clear: the company behind an assistant used by 1.2 billion people a week, according to its spokesperson, is hiring people who know how to plan and book trips with AI agents.
Davidson said Odessia helped early users plan millions of dollars' worth of travel, and that he built it by directing teams of AI agents.
This Is Not A Travel Story. It Is A Transaction Story.
Travel is one of the first places where an assistant stops recommending and starts spending. Once agents plan trips, the next question is who approves the booking, which policy applies and where the receipt goes.
For enterprises, that means your travel policy, your approved suppliers and your expense rules may soon be read by an agent before any employee sees them.
What You Should Prepare For
The wrong reaction is to wait until OpenAI ships a product, then scramble to write rules for it.
The better alternative is to make your own policies machine-readable and owned now, so any agent your people use books inside them.
Write travel and spend rules as data. Limits, approved suppliers and approval steps should live in one governed place, not a PDF.
Decide where agents may pay. Agree whether an agent can complete a purchase or only prepare it for a person.
Keep the audit trail. Every agent-made booking should be traceable to who asked for it and which rule allowed it.
The organizations moving first will not just let AI agents book for their people. They will make sure every booking follows rules the business wrote, owns and can prove.
Journey Towards AGI
Research and advisory firm guiding on the journey to Artificial General Intelligence
Know Your Inference Maximising GenAI impact on performance and Efficiency. | Model Context Protocol Connect with us, and get end-to-end guidance on AI implementation. |
Your opinion matters!
Hope you loved reading our piece of newsletter as much as we had fun writing it.
Share your experience and feedback with us below 'cause we take your critique very critically.
How's your experience? |
Thank you for reading
-Shen & Towards AGI team