Investment memo · No. 003

Agentic AI in 2027: from pilots to production

2026 proved that AI agents can do real work. 2027 will decide which ones survive contact with production. Our outlook, the evidence behind it, and what we look for in founders.

3 min readCash Capital

Our position

In 2026 almost every company experimented with AI agents: software that plans a task, uses tools and takes actions with limited supervision. Far fewer put agents into production at scale. We expect 2027 to be the year that gap closes for some categories and becomes permanent for others. The agents that cross over will be the ones that are measurable, permissioned and priced on the work they complete.

Where adoption really stands

McKinsey's 2026 State of AI survey, fielded in May and June 2026 with 1,719 respondents, found that nearly nine in ten organizations use AI regularly in at least one function. Only about two in ten say they have reached the scaling phase with AI agents across the organization. Among companies with more than $1 billion in revenue the figure is higher, at 40%, up from 27% a year earlier.

The value question is still open. Thirty-seven percent of respondents attribute at least some EBIT impact to AI, essentially unchanged from the year before, and the share of "AI high performers" has stayed flat at about 6%. Adoption is broad; measurable return is concentrated.

Signal2026 readingSource
Organizations using AI regularlyNearly 9 in 10McKinsey, 2026
Scaling AI agents across the organizationAbout 2 in 10 (40% of $1B+ revenue companies)McKinsey, 2026
Attributing some EBIT impact to AI37%McKinsey, 2026
Enterprise apps with task-specific agents by end of 202640% predicted, up from under 5% in 2025Gartner, Aug 2025
Agentic AI projects canceled by end of 2027Over 40% predictedGartner, Jun 2025

Gartner's two predictions frame the year ahead. It expected 40% of enterprise applications to include task-specific agents by the end of 2026, and it also expected more than 40% of agentic AI projects to be canceled by the end of 2027 because of cost, unclear business value or weak risk controls. Both can be true at once, and that is the opportunity: the market is large, and most current attempts will not make it.

What will separate winners in 2027

Measured outcomes, not demos

Buyers have seen enough demos. The agents that get renewed will report accuracy, completion rates and exception rates on the customer's own work, and will improve those numbers over time.

Permissions and identity

An agent that can act needs credentials, scoped access and an audit trail. Enterprises will not let agents touch systems of record without them. Identity, authorization and observability for non-human actors is becoming its own infrastructure layer.

Pricing tied to work

Seat pricing breaks when software does the job instead of helping a person do it. We expect more agent companies to price per task, per resolution or per outcome, and to show gross margins that hold up as model prices change.

Narrow before broad

General-purpose agents attract attention. Vertical agents that own one workflow end to end, such as claims intake, collections, procurement or IT tickets, are reaching production faster because the task is bounded and the value is easy to measure.

Where we see opportunity

  • Vertical agents in regulated and document-heavy work: insurance, financial operations, healthcare administration, legal and government services.
  • Agent infrastructure: identity and permissions for agents, evaluation and testing, observability, and guardrails that let a compliance team say yes.
  • Systems that learn from exceptions: products where every human correction becomes training data, so accuracy compounds with use.
  • Agent-native systems of record: new software built on the assumption that most actions will be taken by agents rather than people.

What we are cautious about

  • Agents that depend on a single model provider's roadmap for their core capability.
  • Products that report usage but cannot show completed work or measured savings.
  • Pricing that assumes today's inference costs will hold, in either direction.

What we want to see from founders

  1. The specific unit of work your agent completes, and how the customer measures it.
  2. Accuracy and exception rates on real customer data, and how they trend.
  3. How your agent is permissioned, audited and stopped when it is wrong.
  4. Why your product gets better with every customer in a way a competitor cannot copy.
Building an agent that does real work? Pitch us. See also our essay Software is becoming labor.

Frequently asked questions

What is agentic AI?

Agentic AI refers to software that can plan a multi-step task, use tools and take actions with limited human supervision, rather than only answering questions.

How many companies have scaled AI agents in 2026?

According to McKinsey's 2026 State of AI survey, about two in ten organizations report scaling AI agents across the organization, rising to 40% among companies with more than $1 billion in revenue.

What does Cash Capital look for in agentic AI startups?

A clearly defined unit of work, measured accuracy on real customer data, strong permissions and audit trails, and a product that improves with every customer.

Sources

  1. McKinsey: The state of AI in 2026
  2. Gartner: 40% of enterprise apps will feature task-specific AI agents by 2026
  3. Gartner prediction on agentic AI project cancellations, via The Star
  4. Crunchbase News: Q3 2026 global startup funding, billion-dollar rounds and exits

Important information. This publication is provided for general information only and reflects the views of Cash Capital as of its date. It is not investment, legal, tax or accounting advice, and it is not an offer to sell, or a solicitation of an offer to buy, any security or any interest in any fund.

Figures, statistics and company information are drawn from third-party sources that we believe to be reliable, including news reports, data providers and company announcements. We have not independently verified them, and some may be estimates, preliminary, reported rather than confirmed, or later revised. Cash Capital makes no representation or warranty as to their accuracy, completeness or timeliness and accepts no liability for any loss arising from reliance on them. Readers should consult the original sources and their own advisers before making any decision.

Predictions, outlooks and other forward-looking statements are opinions about uncertain future events and may not occur. We do not undertake to update this publication. Companies named are used for illustration only; unless stated otherwise they are not Cash Capital portfolio companies, and naming them does not imply any relationship, endorsement or recommendation. See our Terms & Disclosures.

Building something consequential?

Send us your pitch. A partner reads every submission, and you will hear back from us either way.

Pitch Us ›