# Where AI actually lands in a company, and who holds it

> Which parts of a business does AI actually get deployed into first?

Page: https://flyvolo.ai/en/notes/where-ai-actually-lands-in-a-company
Published: 2026-09-13

Adoption is narrower than the coverage suggests and lands in the same three places. The interesting question is not which function buys the tool — it is which role inherits the work the tool creates, because in the verified record that role is almost never the one that chose it.

## What this note does not establish

The adoption figures are US firms only, for a reference period of November 2025 to January 2026, and count firms rather than workplaces or roles. A firm answering yes has AI somewhere in a business function; nobody asked which roles that function contains or whose tasks changed. The duty records are rules in force, not enforcement. And the per-role readings here come from occupation pages whose judgements are labelled task by task — each as evidenced or inferred, and the page says which rather than this one.

## Three functions, and rarely more than three at once

Among US firms that use AI, 57% use it in three business functions or fewer. The most common are sales and marketing (52%), strategy and business development (45%), and IT (41%). Overall 18% of firms use AI in any business function, 32% weighted by employment, expected to reach 22% within six months. For very large firms in information, professional services and finance the rate reaches 50 to 60%.

If your company has AI in more than three functions, you are at the front of this distribution, not behind it. That is the opposite of the impression most coverage gives, and it is worth knowing before you buy anything on the grounds of falling behind.

A second, independent reading points the same way. The Gen AI Jobs Transformation Map that Singapore's central bank commissioned with the sector's skills bodies expects roughly 80% of the country's financial-sector value to come from four business functions — sales and marketing, customer operations, risk management, and engineering and technology. Different country, different method, different decade of regulation, and sales and marketing is at the top of both lists.

Rests on:
- Census BTOS supplement: 18% of US firms used AI in a business function, and 57% of those in three functions or fewer — https://flyvolo.ai/en/changes/ev-20260401-ai-implementation-lead-4 (source: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf)
- Singapore's financial regulator and the sector's skills bodies forecast that Gen AI will augment most financial-sector roles over five years rather than displace them — https://flyvolo.ai/en/changes/ev-20250401-sales-account-manager-3 (source: https://www.ibf.org.sg/docs/default-source/jtm/genai-jtm.pdf)

## What gets replaced first is other software

In the same survey, 16% of AI-using firms replaced existing software or equipment with an AI-integrated version — roughly twice the share reporting any AI-related employment change in either direction. Ninety-five per cent report no headcount change at all.

That lands on a specific person. The process you configured, the rules you wrote and the boundary you set around what the system may do without a human all move to a platform that ships its own versions of them. The verified record does not say anybody lost a job; it says the object of somebody's job was swapped underneath them.

Rests on:
- Census BTOS supplement: 16% of AI-using US firms replaced existing software or equipment with AI-integrated solutions — https://flyvolo.ai/en/changes/ev-20260401-business-systems-owner-3 (source: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf)

## Some of it is already inside without a decision

The same survey finds 23% of firms report workers using AI in work-related tasks, against 18% adopting it at firm level. Diffusion runs both ways: worker use sometimes happens without formal adoption, and formal adoption sometimes happens without worker use.

Vendor telemetry sees the bottom-up half directly, because only the party operating the tool can. Anthropic's usage data shows the median agentic session producing a blog post contains a single human prompt, against thirteen exchanges in a chat conversation — and the gap appears across 26 of 31 output types and survives holding the model constant. Whoever is producing that output inside your company did not raise a purchase order for it.

Rests on:
- Census BTOS supplement: 18% of US firms used AI in a business function, and 57% of those in three functions or fewer — https://flyvolo.ai/en/changes/ev-20260401-ai-implementation-lead-4 (source: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf)
- Anthropic's usage telemetry: the median Claude Code session that produces a blog post contains one human prompt, against 13 rounds of back-and-forth in chat and Cowork — https://flyvolo.ai/en/changes/ev-20260626-copywriter-3 (source: https://www.anthropic.com/research/economic-index-june-2026-report)

## The duties arrive assigned to nobody

From 2 December 2027 the EU AI Act requires deployers of high-risk systems to assign human oversight to natural persons with competence, training and authority. A survey of 200 SRE and DevOps leaders reported that 43% of AI-generated code changes still need manual debugging in production after passing QA and staging. IBM reported its AskHR agent containing 94% of common HR questions across more than 80 automated HR tasks — which moves what is left to people, and what is left is what was already hard.

The recurring shape across these records: capability arrives, a duty arrives with it, and the duty is inherited by whoever already understands both the system and the business consequence. That person usually has the knowledge and not the authority — and the record shows what happens next, because the review step is the first thing cut when it produces no artefact.

Rests on:
- The EU AI Act requires deployers of high-risk AI systems to assign human oversight to natural persons with the necessary competence, training, authority and support (Article 26), from 2 December 2027 — https://flyvolo.ai/en/changes/ev-20260802-ai-implementation-lead-2 (source: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:02024R1689-20260727)
- A survey of 200 SRE and DevOps leaders reported 43% of AI-generated code changes still need manual debugging in production after passing QA and staging — https://flyvolo.ai/en/changes/ev-20260414-software-tester-2 (source: https://venturebeat.com/technology/43-of-ai-generated-code-changes-need-debugging-in-production-survey-finds)
- IBM reported its AskHR agent containing 94% of common HR questions across more than 80 automated HR tasks, with support tickets down 75% since 2016 and HR operating costs down 40% over four years — https://flyvolo.ai/en/changes/ev-20250501-hr-recruiter-3 (source: https://www.ibm.com/case-studies/ibm-askhr)

## The part that argues against us

We sell an assessment of what AI changes for the roles a company employs. The strongest single finding in this note argues that for most companies the answer so far is: not the headcount. Ninety-five per cent of AI-using firms report no employment change. If you came here expecting a number for how many people you will not need, the verified record does not contain one, and we would rather say so than sell you the number somebody else will.

What the record does contain is narrower and more actionable: which functions adopt first, which duties arrive with the adoption, which roles inherit them without authority, and which of your roles the public evidence cannot see at all. That last one is the honest limit of any assessment, ours included, and it is why the empty cell is left visible.

Rests on:
- Census BTOS supplement: 16% of AI-using US firms replaced existing software or equipment with AI-integrated solutions — https://flyvolo.ai/en/changes/ev-20260401-business-systems-owner-3 (source: https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf)
- C.H. Robinson's Q2 2026 10-Q shows average headcount down from 12,858 to 11,471 while revenue rose 19%, naming automation and AI to cut manual processes as part of its first restructuring initiative — https://flyvolo.ai/en/changes/ev-20260630-operations-coordinator-1 (source: https://www.sec.gov/Archives/edgar/data/1043277/000104327726000031/chrw-20260630.htm)

More: https://flyvolo.ai/llms.txt