The News
CData VP of AI Architecture Amit Naik joined the AppDevANGLE podcast to discuss the company’s Connect AI product, a managed MCP (Model Context Protocol) server designed to solve enterprise AI data connectivity challenges. Naik’s central argument is that most AI project failures today are misdiagnosed as model problems when the real culprit is poor context management: feeding models irrelevant, stale, or overly broad data that degrades output quality and inflates token costs. CData positions Connect AI as a governed, managed alternative to the home-built or open-source MCP servers that enterprises are currently assembling under time pressure, with a focus on three pillars: connectivity, context scoping, and access control.
Analyst Take
The Real AI Bottleneck Is Upstream of the Model
The enterprise AI conversation has been dominated for three years by model selection, fine-tuning strategies, and prompt engineering frameworks. CData is making a sharper argument: that era is over. Today’s frontier models are capable enough. The constraint has shifted to data infrastructure, specifically how cleanly and precisely enterprise data reaches the model at inference time. This diagnosis is arriving at the right moment. The industry is beginning to recognize that RAG pipelines, vector databases, and embedding strategies are only as effective as the underlying connectivity layer that feeds them.
What Naik describes as context engineering is not a new problem, but agentic AI makes it a far more consequential one. In a chat interface, bad grounding data produces a wrong answer. In an agentic system with write access to a procurement platform or HR database, bad grounding data triggers a wrong action. The blast radius of a poorly scoped context window is no longer a degraded user experience; it is a corrupted transaction or an unauthorized record update. For ITDMs, this reframes AI connectivity from a developer tooling concern into an enterprise risk management concern that belongs in the same conversation as access controls and audit logging.
Token Economics: The Budget Kill Switch
Naik’s framing of token consumption as a “cut line” for executive AI investment is analytically sharp. The scenario he describes, a company exhausting its annual AI budget within three months because every conversation carries an oversized context, is not hypothetical. It is the natural outcome of bolting AI onto analytic data stores without first solving the scoping problem. Padding a context window with 120 Salesforce fields when the use case requires 10 is not just inefficient; it forces the model to spend compute cycles parsing irrelevance rather than reasoning over signal.
For developers, the practical implication is architectural. Real-time, scoped data access through a well-governed connector layer is not a nice-to-have optimization. It is the difference between an AI project that survives its first budget review and one that gets cancelled on cost grounds before it reaches production scale.
Governance Is the Moat, Not the Feature List
The most strategically important part of Naik’s argument is his treatment of open-source and vibe-coded MCP servers. The MCP ecosystem is moving fast, and the OWASP Top Ten list for MCP vulnerabilities, including confused deputy attacks and malicious tool injection, represents real attack surface that most enterprise security teams have not yet audited. According to ECI Research’s Google GovTech Survey, 31.8% of respondents say that FedRAMP and compliance approval friction is the single largest blocker preventing widespread AI adoption in developer workflows. That friction exists for a reason, and ad-hoc MCP implementations built by individual teams are unlikely to satisfy the audit trail requirements that compliance officers will eventually demand.
This is where CData’s positioning as a managed, governed MCP server carries differentiation potential, provided the company can demonstrate auditability and permission granularity at the level enterprise security architects require. The governance angle also maps to a broader structural reality in how government and regulated-industry teams are deploying generative AI tools. ECI Research’s Google GovTech Survey found that 36.6% of respondents are deploying generative AI through isolated GovCloud environments, and another 27.6% are using self-managed cloud deployments within their own infrastructure. Neither of those deployment patterns is well-served by multi-tenant commercial SaaS AI connectors. CData’s architecture, which supports scoped, policy-governed access rather than broad data dumps, aligns with where the regulated-market segment is heading.
There is also an agent identity question that Naik raises: when an agent acts on behalf of a person, or on behalf of another agent in an A2A scenario, attribution becomes genuinely difficult. Mapping agent identity back to a human principal for compliance purposes is a hard problem, and the market does not yet have consensus on how to solve it. CData’s governance framing positions them to address this, but the product would need to demonstrate concrete identity attribution and audit logging capabilities to satisfy regulated-industry buyers.
The data point that should concern ITDMs most: according to ECI Research’s Google GovTech Survey, 48.0% of respondents identify navigating compliance documentation and audit evidence collection as the greatest source of cognitive load for their developers today. Adding poorly governed AI data connectivity to that environment does not reduce cognitive load; it multiplies it.
Looking Ahead
CData is entering the MCP server market at exactly the right time. The protocol is gaining adoption fast enough that enterprises are making procurement decisions now, often defaulting to open-source implementations because managed alternatives are still limited. The window for a vendor like CData to establish itself as the default governed connectivity layer is open, but it will not stay open long. Hyperscalers, major integration platform vendors, and API gateway providers are all circling the same problem.
The deeper market shift to watch is the convergence of AI connectivity governance with broader DevSecOps and zero-trust architectural mandates. As agentic systems proliferate, the question of what data an agent can access, under what identity, with what audit trail, will become a standard line item in security reviews. Vendors that can answer those questions with production-grade tooling rather than architectural promises will win the category.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.