Fresh context for agents. Rebuilt every 12 hours. Provenance by Synorb.
Updated 2026-11-20 12:00 UTCnext update 2026-11-21 00:00 UTC
Confirmed · 3 Manifests883 words

Two in three enterprises now depend on AI — but only a third have a plan to run it

Two in three enterprises would break if AI stopped, but only a third have a formal operating model. The gap is organisational, not technical — and business units are moving anyway.

Key facts
  • A study sponsored by CoreWeave and conducted by Humatica finds 66% of enterprises run processes or business models that would break if AI were disabled, while only 35% have a formal operating model for AI. Self-reported m:…5144
  • The same study finds 42% of production AI use cases started in business units rather than IT, 24% were started by an employee or team on their own initiative, and leaders cite their own organisation as a constraint 2.4 times as often as the technology. Self-reported m:…5144
  • Quantiphi, a Google Cloud partner, says enterprise AI must connect securely to core systems like electronic health records and banking engines, and reports implementing over 100 enterprise agents in 100 days. Self-reported m:…0583
  • Google reports small businesses using Gemini in Workspace, including Bay Area Kids Rentals closing over 50% of incoming leads, Anderson Foundries turning manuals into training resources, and Sourmilk managing production in Google Sheets. Self-reported m:…5919

Dependence without a plan

A new study sponsored by CoreWeave and conducted by Humatica, titled AI-Native by Design: Operating Model Behind Lasting AI Advantage, puts a number on a situation many executives already feel. It finds that 66% of enterprises in the study run processes or business models that would break if AI were disabled, and that only 35% of organisations have implemented a formal operating model for AI. The gap between those two numbers is the story. AI has become load-bearing in two-thirds of the enterprises surveyed, and the structures for running it — who owns it, who secures it, who pays for it when it breaks — exist in only a third. The study was sponsored by a company that sells AI infrastructure, so its numbers should be read with that in mind, but the shape of the finding matches what the rest of this week's evidence shows: adoption has outrun organisation. m:…5144

The study also explains why the gap is hard to close from the centre. It finds that 42% of production AI use cases started in business units rather than IT or central AI teams, and nearly one in four, 24%, was started by an employee or team on their own initiative. Leaders cite their own organisation as a constraint on AI 2.4 times as often as the technology itself. This suggests the operating model problem is not a shortage of models or tools. It is that the people using AI are distributed across the business, the decisions about it are being made locally, and the central team is trying to govern something that has already spread. The implication is uncomfortable for anyone who likes orderly rollouts: the rollout has already happened, and it happened without you. m:…5144

What the infrastructure argument is

The study's sponsors have an answer, and it is worth hearing even while discounting the source. It finds that 73% of enterprise and investment leaders cite differentiation and proprietary intellectual property as the primary reason to move workloads to controlled infrastructure, which is the case for owning the stack rather than renting it by the call. CoreWeave describes its own offerings in the same release, including a Zero Trust AI Framework with six layers of control over sensitive AI workloads. These are vendor claims, and they should be treated as such, but the underlying argument is one that executives should weigh on its merits: if AI is now load-bearing, the question of who controls the infrastructure it runs on is a strategic one, not a procurement detail. m:…5144

Quantiphi, an official launch partner for Google Cloud's industry-specific Gemini capabilities, makes a similar argument from the services side. The firm says successful enterprise adoption requires grounding agents in verified institutional data and connecting them securely to core systems like electronic health records, core banking engines, and claims management platforms, and it reports implementing over 100 enterprise agents in 100 days using repeatable, governed patterns. The emphasis, again, is on connection rather than capability: the hard part is not getting an agent to answer, but getting it to answer from your systems, under your rules, with your data. m:…0583

The small-business version

Google's account of small businesses using Gemini in Workspace shows the same pattern at the other end of the scale, and it is useful precisely because the stakes are smaller and the results are easier to see. Bay Area Kids Rentals reported closing over 50% of incoming leads and doubling revenue after using Gemini in Gmail for logistics and inquiry management, including daily to-do briefs and pricing retrieval. The company uses Gemini in Gmail to draft replies, organise to-do lists, and manage customer inquiries. These are Google's own examples, reported by the company, and they should be read as such, but they illustrate the point the enterprise study makes: the value is in the daily work, not in a showcase project. m:…5919

The other examples follow the same shape. Anderson Foundries uses Gemini Notebook to turn technical manuals into interactive training resources, and the company says it saves thousands of dollars by drafting quality manuals in-house with Gemini. Sourmilk uses Gemini in Google Sheets to create production models and manage email triage, and scaled into more than 100 grocery stores within its first year as a two-person team. Curious Elixirs uses Gemini in Google Docs for marketing campaigns and to source ingredient suppliers. None of these are dramatic AI stories, and that is the point. The technology is being absorbed into ordinary operations, and the organisations that benefit are the ones that let it into the daily work rather than keeping it in a pilot. m:…5919

What to do about the gap

The decision facing executives is not whether to adopt AI. That decision has already been made, by their own employees, in their own business units, and the study's numbers say the dependency is already there. The decision is whether to fund the operating model now, while the gap is visible, or to keep running load-bearing systems on informal arrangements. The honest answer depends on how much of your business would actually break if AI stopped, and the study's 66% figure is a prompt to measure that rather than to guess. If the number is anywhere near that for your organisation, the operating model is not overhead. It is the thing that keeps the dependency from becoming a liability the first time something breaks, leaks, or needs an owner. m:…5144 m:…0583

What is still unresolved

  • How many enterprises were surveyed, and whether the 66% figure holds beyond the study's sample
  • What a formal AI operating model actually costs to build, in time and headcount
  • Whether business-unit-led adoption produces better or worse outcomes than centrally governed rollouts

Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.

Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.

Run these Manifests yourself

Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them.

  1. Get a Synorb Key. Click Generate Free Agent Key at https://synorb.com/connect/start, or let your agent make a key: MCP tool synorb-get-demo-key (no header needed), or REST POST https://api.synorb.com/api/connect/demo-key with a JSON body such as {"client":"your-agent-name"}. The key is the api_key field of the answer. Save the key. Existing 72 Hour Agent Demo keys keep their issued limits. New keys require email OTP verification. An existing demo key lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; all of them are starter Beacons.
  2. Read each cited Manifest by id. REST: GET https://api.synorb.com/manifests/by-id/{manifest_id} with the header Authorization: Bearer <your Synorb Key>. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief.
  3. Check the claim. Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them.
Cited Manifest ids (3)
  • 1795093383266725144
  • 1795113007295840583
  • 1795109919257805919
for id in 1795093383266725144 1795113007295840583 1795109919257805919; do
  curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id"
done

Provenance ledger

ManifestTitleSourceClaims usedPublished
1795093383266725144Humatica Study Finds Two in Three Enterprises Now Depend on AI, But Operating Models Are Still Catching upcoreweave-ir-news1795093633966726908, 1795093633967726909, 1795093633968726910, 1795093633970726913, 1795093633970726912, 1795093633964726907, 179509363396972691111-19 00:00
1795113007295840583From AI Experiments to Industry Execution with Geminiquantiphi-blog1795113416850844033, 1795113416851844034, 1795113416846844032, 179511341685384403511-19 00:00
1795109919257805919Small businesses, big results: How teams scale with AIgoogle-cloud1795110118725806917, 1795110118726806918, 1795110118728806920, 1795110118721806914, 1795110118724806916, 1795110118727806919, 1795110118729806921, 1795110118730806922, 179511011873180692311-19 00:00

Related stories