Nylas last month published The State of Agentic AI 2026, a survey-based report drawing on responses from 1,026 developers, product managers, and technology leaders collected between December 18 and December 30, 2025. The findings document a technology category that has moved well past the experimental stage inside organizations while still carrying the public image of something new and unproven. For marketing technology professionals, the numbers signal a fast-approaching inflection point.

The survey was conducted with the assistance of Centiment. Respondents were primarily based in the United States and worked across engineering, product management, IT, data, and executive leadership roles. According to the report, nearly 80% of respondents were directly involved in building or influencing agentic AI decisions at their organizations. Most worked at mid-to-large companies, with a strong concentration in software and SaaS, and meaningful representation from finance, healthcare, consulting, and customer experience sectors.

What the numbers actually say

The headline figure is direct: 48% of respondents said they are actively building agentic AI today. Another 31% described themselves as seriously evaluating it. Only 5.7% said they were not building agentic AI at all. That distribution leaves little room for the narrative that the industry is still watching from the sidelines.

Equally telling is the roadmap data. According to the report, 64.4% of respondents already have agentic AI on their product roadmap, with another 29% planning to add it within 12 months. Only 6.5% said no. More than half - 55.3% - described it as very important to their overall product strategy, and a further 17.4% called it critical. Fewer than 3% considered it irrelevant.

The timeline for market normalization was just as stark. When asked when agentic AI would become table stakes, 36.3% said within 12 months, and 49.1% said within two to three years. Taken together, 85% of respondents believe the technology will be an expected baseline capability within three years. Only 4.8% said never.

No shared definition, significant shared behavior

Perhaps the most counterintuitive finding in the report concerns terminology. Despite the intensity of deployment, there is no consensus on what agentic AI actually means. Asked to choose a definition, respondents split across multiple options simultaneously. According to the report, 58.6% described it as workflow engines that trigger actions across systems. Another 51.8% said autonomous background workers. Chat-based assistants gathered 46.7%, and automated background jobs or schedulers drew 45.6%. No single interpretation dominated.

The definitional fragmentation extends to organizational level. When asked which description best fit their current use of agentic AI, 58.7% chose autonomous systems that execute complex workflows without human intervention. But 50.4% simultaneously selected AI assistants that recommend actions but require human approval. Only 8.9% said they had no clear definition yet.

The report frames this ambiguity as a feature rather than a flaw. According to the authors, the category is still being shaped in real time by what teams are building, testing, and slowly learning to trust in production. The definition that sticks will be the one attached to the first workflows teams cannot imagine turning off.

Internal workflows first, customer-facing products second

The data on deployment context reveals a deliberate sequencing. According to the report, 74.1% of respondents currently use AI tools for internal productivity, and 66.2% use them for internal operations. Embedded customer-facing deployment trails at 38.8%. This gap is not accidental.

Internal environments offer lower risk thresholds. Engineering automation, IT workflows, operational tooling - these are places where teams can move quickly, break things, and learn without consequences visible to customers or regulators. The pattern is consistent across industries and company sizes. Teams earn confidence internally, then expand outward.

The workflows where agentic AI has proven value already reflect that sequencing. According to the report, IT and internal operations led all categories at 76.8%. Customer support followed at 58.1%, sales workflows at 53.5%, and project and delivery management at 49.4%. Finance and reporting reached 33.8%, recruiting and HR 30.1%, and meeting follow-ups and documentation 35.6%.

That ranking mirrors the friction profile of these workflows. IT operations and customer support involve repetitive, time-sensitive tasks spread across many systems, exactly the conditions where autonomous execution removes the most friction. Customer-facing deployments require higher trust thresholds and more governance infrastructure before teams feel comfortable letting agents act.

For marketing technology teams in particular, the advertising industry's agentic AI deployments through late 2025followed this same internal-first pattern. Platforms that spent the fall testing agentic capabilities on internal campaign workflows entered early 2026 with production systems already running.

The trust architecture underneath adoption

Full autonomy remains rare. According to the report, only 4% of teams allow agents to act without any human approval - what the survey labeled "YOLO mode." The overwhelming majority operate in more cautious configurations. Intermediate mode, where agents handle some actions automatically and others require approval, accounted for 52.6% of respondents. Approval mode, requiring human sign-off on all agent actions, represented 38.1%.

This graduated approach reflects a pragmatic calculation, not philosophical hesitation. Teams extend autonomy for specific task categories where the cost of a mistake is low and the value of speed is high. According to the report, 64.9% of respondents said they would trust an agent to send internal reminders without approval. Creating tickets or tasks drew 60.9%. Scheduling meetings reached 54.6%, sending customer emails 49.1%, and updating internal records 45.0%. Changing system settings received far less trust at 29.9%.

The pattern is consistent with how the advertising industry has approached agentic AI governance frameworks - autonomy calibrated to risk level, with human oversight preserved for higher-stakes decisions. Agents earn expanded permissions through demonstrated reliability on lower-stakes tasks, not through theoretical capability alone.

This graduated trust model has a formal name in the Nylas glossary. It describes autonomy as existing on a spectrum, from fully human-approved actions to fully autonomous execution, with most production systems operating somewhere in between. The technical term for AI systems that plan and execute multi-step tasks, use tools and APIs, maintain context, and adapt based on outcomes is agentic AI. The practical reality is that most systems described this way still include a human somewhere in the approval chain.

Speed matters more than cost

The report provides a clear answer to why teams are investing in agentic capabilities: velocity. Improving response time and speed ranked first among adoption motivations at 69.3%. Reducing manual work followed at 56.2%, and improving data quality across systems at 55.6%. Reducing costs came in at 39.0%. Meeting leadership expectations trailed at 13.5%.

That ordering carries implications for how marketing technology vendors should frame agentic capabilities. Arguments centered on cost savings resonate less than arguments about operational speed. For organizations that have watched MiQ's research showing 72% of marketers plan to expand AI use but only 45% feel confident implementing it, the velocity framing addresses a concrete problem: the pace of execution is lagging behind the ambition to deploy.

Speed as a motivation also helps explain where adoption concentrates. IT operations and customer support are precisely the functions where delays in handoffs and coordination create the most visible inefficiencies. When agents reduce response latency on a support ticket or automate the creation of follow-up tasks, the value is immediate and measurable. The cost savings calculation is harder to run. The speed improvement is visible from day one.

What is actually slowing teams down

The barriers to deployment are technical and organizational in roughly equal measure. According to the report, security approvals topped the list of implementation blockers at 49.8%. Integration complexity followed at 44.1%, engineering bandwidth at 33.5%, unclear ROI at 30.8%, organizational hesitation at 29.2%, and fear of vendor lock-in at 19.6%.

On the technical side, the hard problems are specific. Reliability and failure handling led at 47.6%. Cross-platform integration reached 43.4%. Compliance requirements and workflow orchestration tied at 38.8%. Permissions and identity management followed at 37.8%, and monitoring and observability at 30.8%.

The distribution across these categories points to a common theme: agents that work in demos frequently break in production. Platform changes and API updates introduce instability. Permissions management becomes complicated when agents act across multiple systems on behalf of users. Audit trails are difficult to maintain. These are not problems that better prompting solves. They require infrastructure investment, engineering resources, and organizational processes.

That infrastructure gap is relevant context for the data governance problems documented elsewhere in the industry. The Publicis Sapient research released in late 2025 found that most organizations remain in pilot mode despite claiming enterprise readiness. The Nylas data suggests the specific blockers are granular and solvable - security approvals, integration plumbing, reliability engineering - rather than fundamental capability gaps.

Agentic AI as a vendor switching trigger

The vendor landscape implications of the report deserve close attention. According to the findings, 94% of respondents said they would possibly or very likely switch vendors if a competitor offered stronger agentic AI capabilities. That breaks down to 46.1% very likely and 47.5% possibly. Only 4.5% said unlikely, and 1.9% said no.

That figure transforms agentic AI from a feature into a retention and acquisition variable. Vendors that fail to keep pace risk losing accounts to competitors that do, regardless of the depth of the existing relationship. The switching event, as the report labels it, is the moment when a capability gap becomes large enough to outweigh the friction of migrating to a new platform.

The dynamics mirror what PPC Land documented in October 2025 around the Ad Context Protocol, where the advertising industry began weighing whether standardized agentic interfaces could reshape programmatic relationships. When agents become buyers and platform capabilities become machine-readable requirements, the cost of switching drops and the pressure on lagging vendors rises.

Reliability, integration coverage, and security drive the switching calculus more than novelty. According to the report, teams are looking for agentic systems they can trust to run inside real workflows, not just perform well in sales demonstrations. The question that matters is whether an agent can run every day without manual intervention and integrate cleanly with the rest of the stack. Vendors that can answer yes to both are acquiring leverage. Those that cannot are accumulating churn risk.

What this means for marketing technology

The marketing community has been watching agentic AI deployments with a mixture of urgency and caution throughout late 2025 and into early 2026. Amazon launched its Ads Agent in November 2025 for automated campaign management across Amazon Marketing Cloud and Amazon DSP. LiveRamp introduced agentic orchestration in October 2025, giving autonomous agents governed access to its identity resolution and activation platform. IAB Tech Lab published its agentic standards roadmap in January 2026 to prevent protocol fragmentation before competing implementations could splinter the ecosystem.

The Nylas data provides the demand-side perspective on all of that activity. The reason platforms are investing in agentic infrastructure is that the teams building software on top of them are already doing the same. Nearly two-thirds of respondents have agentic AI on their product roadmaps. More than half consider it critical or very important to their overall strategy. And almost all of them are willing to switch vendors if a better agentic option becomes available.

The marketing technology implications are direct. Campaign automation that requires constant human supervision will lose ground to campaigns that run autonomously within defined parameters. Platforms that offer graduated trust controls - letting agents handle low-risk decisions while preserving human oversight for high-stakes ones - will attract more adoption than those offering only binary options. And vendors that invest in reliability, compliance, and cross-platform integration will retain accounts that might otherwise migrate.

The 85% figure - the share of respondents who believe agentic AI will be table stakes within three years - is not a prediction about distant technology futures. For the majority of respondents, that tipping point arrives within 12 months. The teams building products for marketing professionals are already planning around it.

Timeline

Summary

Who: Nylas, a communications infrastructure company, commissioned the survey and published the report. Respondents were 1,026 developers, product managers, IT professionals, and technology executives primarily based in the United States, with 78.5% directly involved in building or influencing agentic AI decisions at their organizations.

What: "The State of Agentic AI 2026" documents the current state of agentic AI adoption across the software industry, covering definitions, deployment patterns, trust models, adoption motivations, implementation barriers, and vendor switching intentions. Key findings include 48% of respondents actively building agentic AI, 85% expecting it to become table stakes within three years, and 94% willing to switch vendors for stronger agentic capabilities.

When: The survey was conducted between December 18 and December 30, 2025, with the report published today, March 10, 2026.

Where: The survey covered respondents primarily in the United States, working across mid-to-large organizations in software, SaaS, finance, healthcare, consulting, and customer experience sectors. Deployments documented in the report are predominantly internal, with IT operations and customer support leading adoption patterns.

Why: The report addresses a gap between public perception of agentic AI as experimental and the operational reality inside organizations where production deployments are already running. The findings are intended to help developers and product leaders understand where the market is heading, what is holding teams back, and how agentic capabilities are reshaping vendor relationships across the technology industry.

Share this article
The link has been copied!