Enterprise AI agent orchestration is consolidating around major model-provider platforms, led by Anthropic’s Claude, yet most deployed “agents” remain simple chat interfaces rather than true multi-step systems — a gap with direct implications for cryptocurrency, blockchain, and digital-asset operations that depend on reliable, auditable automation.

AI Integration

A new VentureBeat Pulse Research wave, fielded in June 2026 among 101 organizations with 100 or more employees, finds that orchestration decisions are concentrating on the platforms tied to leading models. Anthropic’s Claude is named the primary platform by 40% of respondents, more than double any rival. Microsoft AI Foundry / Copilot Studio follows at 18%, OpenAI’s Agents SDK / Responses API at 13%, and Google’s Enterprise Agent Platform at 8%, with a further 2% on Amazon Bedrock Agents. Open frameworks (LangChain / LangGraph) account for 6%, while 5% build custom in-house and 3% are not yet orchestrating.

The selection logic centers on “model gravity” — native alignment with a state-of-the-art base model (21%). Flexibility across models and tools (17%) and ease of development (17%) closely follow, with security and permissions (14%) and total cost of ownership (11%) rounding out priorities. Performance metrics, such as latency and memory, trail well behind at 4%. Respondents rate the platforms they use at 3.94 out of 5 overall — identical to “value for money” — with “ease of implementation” somewhat lower at 3.85. Notably, 96% say they plan to change their orchestration approach within the next year, signaling provisional acceptance rather than long-term satisfaction.

Technology Use Case

What enterprises optimize for is clear: reliable multi-step execution. Task completion reliability (32%) and multi-step workflow management (28%) together account for the majority of success criteria, ahead of developer productivity (17%), end-user experience (9%), and operational stability (9%). In financial and blockchain contexts — where action chains can include onboarding checks, transaction preparation, risk controls, and settlement — dependable coordination across steps is a practical requirement rather than a theoretical goal.

Yet the report’s central tension is that ambition outpaces reality. When asked to classify their own portfolios, 62% say only 1–25% of their deployed “agents” are genuinely orchestrated; 19% place that share at 26–50%. A further 9% say 0% — every deployment is a chatbot or prompt wrapper — and only 10% of organizations have crossed the halfway mark into orchestrated territory, with 7% reporting 51–75% and 3% at 76–100%. For crypto-market operations that rely on chain-of-custody and compliance-minded automation, this shortfall underscores why simple assistants cannot yet shoulder the core workload.

Market Impact

Strategy is coalescing around three moves over the next 12 months: 25% will increase investment in custom, in-house orchestration control planes; 24% will standardize on a single centralized framework; and 23% will expand agents from sandbox into production. Together, these point to consolidation and productionization. Only 9% anticipate shifting toward turnkey, natively embedded architectures, and 8% expect either model-native autonomy or external frameworks, split evenly. Just 4% expect no change. For trading, custody, and blockchain infrastructure teams, this tilt toward owned control planes reflects the need to codify business rules, permissions, and audit paths within the orchestration itself.

Investment priorities reinforce that push. Agent workflow tooling leads expected budget growth at 34%, followed by security and permissions enforcement at 25% and infrastructure for scaling agents at 20%. Monitoring and debugging draw 11%, with another 11% reporting flat budgets. This spend pattern suggests enterprises are building and hardening orchestration pipelines — the connective tissue for multi-step work — rather than merely instrumenting them. In digital-asset settings, where permissions and segregation of duties are central, the emphasis on policy enforcement tooling is directly pertinent.

Industry Response

Control-plane architecture is where strategy and risk converge. By the end of 2026, a majority (51%) expect a hybrid control plane that combines provider-native capabilities with external orchestration. Another 22% anticipate a custom in-house control plane, and 15% expect external platforms abstracted from model providers. Only 6% foresee handing control to a provider-managed agent service, and 6% do not expect to deploy autonomous agents at scale. Across these options, 88% maintain at least partial control outside a model provider.

The rationale is explicit: vendor lock-in is the top risk when control resides within a provider platform (35%), followed by security and permissioning limitations (28%) and inflexibility across models and tools (21%). Compared to an April–May snapshot (n=145), the June read points toward keeping control: hybrid expectations rose from 34% to 51%, while intent to rely fully on a provider-managed service fell from 12% to 6%. Concern ordering also shifted, with lock-in overtaking security and permissioning as the leading issue. For blockchain and digital-asset organizations, which often need to prove independence of controls, the preference for hybrid governance mirrors established risk practices.

Fiscal Control

Cost governance remains reactive for many. More than a quarter (27%) report that they have only after-the-fact monitoring and no real-time kill switch to halt runaway consumption. Another 32% rely on native platform controls such as budget caps and throttling. A more engineering-led minority builds custom gateway plumbing (23%) to intercept runaway runs, while 19% pursue dynamic routing arbitrage to shift heavy work to lower-cost models. In crypto-facing settings — where “token” can be ambiguous — the report’s “token consumption” refers to AI model usage, not blockchain assets, and the stated practices highlight how few teams can programmatically halt spend before bills accrue.

Methodology and Scope

The findings draw on a single June 2026 survey wave of 101 qualified respondents from organizations with 100 or more employees. The sample spans enterprise-size bands evenly across 100–499, 2,500–9,999, and 50,000+ employees (21% each), with 10,000–49,999 and 500–2,499 at 19% each. Roles include product and program managers (15%), CIO/CTO/CISO (13%), consultants and advisors (13%), and a range of data, AI, and engineering leaders; 81% are recommenders, influencers, or final decision-makers for AI solutions. Industry mix is led by Technology/Software (44%), followed by Financial Services (17%) and Healthcare/Life Sciences (8%). Because this is a single-wave, self-selected sample, results read directionally and are not a probability sample or a spend-weighted market share.

Bottom Line

Enterprises are building on model-provider platforms — with Anthropic’s Claude out in front — and judging orchestration by its ability to carry multi-step work to completion. Budgets favor workflow tooling and permissions, control is trending hybrid to hedge against lock-in, and production exposure is set to rise. Yet most portfolios are still dominated by chatbot wrappers, and fiscal controls over model-token spend are often reactive. For crypto and blockchain teams seeking dependable, governed automation, the report’s message is straightforward: the orchestration layer is taking shape ahead of the truly orchestrated agents it is intended to run.