--- title: AI agents are ready. Your data infrastructure isn't. description: Three separate reports converge on the same finding: agentic AI adoption is being throttled not by model quality but by legacy data infrastructure. A Google... canonical: https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AI agents are ready. Your data infrastructure isn't. | daily.dev og:description: Three separate reports converge on the same finding: agentic AI adoption is being throttled not by model quality but by legacy data infrastructure. A Google... og:url: https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w og:image: https://api.daily.dev/og/posts/QOADX2O4W.png og:image:alt: AI agents are ready. Your data infrastructure isn't. og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AI agents are ready. Your data infrastructure isn't. **[Trends](https://daily.dev/sources/trends)** · 2 min read · 1 upvotes · 0 comments ## Summary Three separate reports converge on the same finding: agentic AI adoption is being throttled not by model quality but by legacy data infrastructure. A Google Cloud/MIT Technology Review survey of 300 executives found only 10% run AI agents at scale, though 69% plan to within two years, with 55% blaming legacy data systems, silos, and slow batch pipelines. Data access correlates directly with trust — orgs giving agents access to over 70% of enterprise data report 100% trust in AI accuracy, versus 22% for limited-access orgs. A related Google Cloud infrastructure report highlights security as a major blocker, with 79% citing security/governance/operations as their top scaling challenge and 35% flagging insufficient security for multi-system access. New attack surfaces like tool poisoning and indirect prompt injection are pushing organizations toward dynamic permissions and human-in-the-loop controls rather than blanket access blocks. The consistent advice: fix the highest-friction data and governance gaps first rather than trying to overhaul everything at once. ## Content There's a pattern forming across every serious analysis of enterprise AI agent deployments right now, and it's not flattering. Gartner predicts over 40% of agentic AI projects get canceled by end of 2027. A 2025 MIT NANDA study puts the generative AI pilot failure rate at around 95%. McKinsey finds 88% of organizations use AI but only 6% qualify as high performers. The model isn't what's failing. The plumbing is. The diagnosis is consistent across sources: agents are hitting legacy data systems that were never designed for them. Transactional databases built for apps, data lakes built for analysts, neither built for probabilistic AI that needs clean, real-time, structured access. Google Cloud and MIT Technology Review surveyed 300 executives and found 55% cite legacy data systems as the main blocker. Organizations giving AI access to over 70% of their enterprise data report 100% trust in AI accuracy. Those with limited access: 22%. The API layer is just as bad. Of 27,310 API providers analyzed by the API Evangelist Rating System, only 771 document idempotency. Only about 1.5% score high enough to be considered "Agent-Native." MCP servers are shipping with long-lived, over-privileged API keys instead of proper OAuth and scoped tokens. Google's Andi Gutmans puts it plainly: the model is rarely the bottleneck anymore. The harder problem is minimizing context to get reliable outcomes at the lowest cost. Security is its own mess. Zero trust architecture assumes known identities and bounded permissions. Agents chain individually-authorized actions into unauthorized outcomes, spawn subagents with no registered identity, and can be hijacked through prompt injection to leak data across agent-to-agent calls that nobody can currently inspect. Docker's framing is useful here: per-harness guardrails are inherently unreliable because agents can talk past them. You need enforcement at the runtime layer, below all the models and harnesses, or you don't really have governance. The practical upshot, if you're deciding where to spend: fix contracts, auth, error handling, and data access before you build MCP servers or agent skills. Add hard timeouts, circuit breaking, and durable checkpointing to anything running in production. Treat model output as untrusted input. Build evaluation datasets before you start tweaking prompts. The frontier models are now within roughly 5% of each other on benchmarks, per Stanford's 2026 AI Index. The competitive differentiator has shifted. It's infrastructure now, and most organizations aren't ready for that conversation. ## Questions this post answers ### What percentage of agentic AI projects will be canceled by 2027 according to Gartner? Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027. This follows a broader pattern of enterprise AI deployment struggles, including an MIT NANDA study putting the generative AI pilot failure rate around 95%, and McKinsey finding that while 88% of organizations use AI, only 6% qualify as high performers. _Engineering leaders weighing agentic AI investment can follow infrastructure readiness debates on daily.dev._ ### Why do AI agents fail even when the underlying model performs well? Agents fail primarily because of legacy data infrastructure and API gaps, not model quality. Transactional databases and data lakes weren't built for the clean, real-time, structured access probabilistic AI needs; 55% of 300 surveyed executives cited legacy data systems as the main blocker, and organizations with access to over 70% of enterprise data reported 100% trust in AI accuracy versus 22% for those with limited access. _Teams debugging unreliable agent behavior can track data-infrastructure fixes discussed around AI agents on daily.dev._ ### How many API providers properly document idempotency for AI agent use? Only 771 of 27,310 API providers analyzed by the API Evangelist Rating System document idempotency, and only about 1.5% score high enough to be considered agent-native. Many MCP servers also ship with long-lived, over-privileged API keys instead of proper OAuth and scoped tokens, undermining reliable and secure agent-to-API integration. _Developers building MCP servers or agent skills can weigh API and auth readiness discussions on daily.dev._ ## Similar posts on daily.dev - [Why 40% of AI Agents Might Fail \(and How To Save Yours\)](https://daily.dev/posts/why-40-of-ai-agents-might-fail-and-how-to-save-yours--qnck3ryb4) · The New Stack · 2 upvotes · 0 comments - [How to stop holding AI agents back](https://daily.dev/posts/how-to-stop-holding-ai-agents-back-86grinkth) · The Next Web · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#gcp](https://daily.dev/tags/gcp), [#big-data](https://daily.dev/tags/big-data), [#prompt-injection](https://daily.dev/tags/prompt-injection), [#ai-governance](https://daily.dev/tags/ai-governance) [View this post on daily.dev](https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AI agents are ready. Your data infrastructure isn't.","url":"https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w"},"datePublished":"2026-08-27T19:35:35.312Z","dateModified":"2026-09-03T16:27:19.400Z","description":"Three separate reports converge on the same finding: agentic AI adoption is being throttled not by model quality but by legacy data infrastructure. A Google...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c91c975eeddf40eae2ca6c0b0a12b9e4?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c91c975eeddf40eae2ca6c0b0a12b9e4?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,gcp,big-data,prompt-injection,ai-governance","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"AI agents are ready. Your data infrastructure isn't."}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ai-agents-are-ready-your-data-infrastructure-isn-t--qoadx2o4w#faq","mainEntity":[{"@type":"Question","name":"What percentage of organizations are actually running AI agents at scale right now?","acceptedAnswer":{"@type":"Answer","text":"Only 10% of organizations are running AI agents at any real scale, according to a Google Cloud and MIT Technology Review survey of 300 data and tech executives. However, 69% plan to scale agent deployments within two years, with legacy data infrastructure cited by 55% of executives as their primary blocker to getting there. Track how data infrastructure shapes agentic AI rollouts by following coverage on daily.dev."}},{"@type":"Question","name":"Does giving AI agents access to more enterprise data actually improve trust in their accuracy?","acceptedAnswer":{"@type":"Answer","text":"Yes, there is a stark correlation: organizations that give AI access to more than 70% of their enterprise data report 100% trust in AI accuracy, compared to just 22% trust among organizations with limited data access. This suggests trust issues with AI outputs often stem from data access gaps rather than model quality. Developers weighing data access strategy for AI agents can follow this debate on daily.dev."}},{"@type":"Question","name":"What new security risks do AI agents introduce compared to traditional applications?","acceptedAnswer":{"@type":"Answer","text":"AI agents introduce attack surfaces traditional security models were not built for, including tool poisoning and indirect prompt injection, plus agents acting across systems with permissions never designed to be delegated. A Google Cloud report found 79% of tech leaders cite security, governance, or operations as their top AI scaling challenge, and 35% specifically flag insufficient security for multi-system agent access. Teams securing agent-to-system permissions can keep up with these emerging risks on daily.dev."}}]} ```