Coder Legion - Recent posts tagged aireadiness

The Purchase Order Is Canceled — and Executed. That's Not a Data Error.

Wed, 16 Sep 2026 12:30:00 +0000

An AI agent trying to answer "what did we invoice this customer" runs into a problem most demos never show: the honest answer isn't one number. There's what you invoiced in U.S. dollars, what that converts to in local currency, and the tax implication layered on top. There's a purchase order that got canceled and a separate one that got executed for the same order. To a system built to give a clean answer, that looks like corrupted data. To Naveen Neelakantam, product and technology strategy lead in Everpure's Office of the CTO, it's just what business data actually looks like — and building AI that tries to erase it is where a lot of agentic pilots start quietly failing.

"Contradiction is part of the data landscape," Neelakantam said in an interview about Everpure's approach to AI-ready data. "Those contradictions are meaningful. You might see a canceled purchase order plus the executed purchase order — we want to link all of those in the knowledge graph and say, if you're asking about POs, some of them are canceled, some are executed."

That's a deliberately different posture from a lot of "clean your data for AI" pitches, which treat conflicting records as noise to filter out before a model ever sees them. Everpure's approach, built on its Everpure Data Intelligence platform — acquired via 1touch.io, and already running in production for financial and insurance customers on-premises, not a fourth-quarter beta, per Neelakantam — keeps the mess and surfaces it instead.

The mechanics are worth understanding if you're the one building the agent. Data Intelligence scans structured sources like Snowflake, Salesforce, and SAP alongside unstructured ones — invoices, scanned purchase orders, object stores — looking for entities: names, addresses, PO numbers. "In these invoices or this OCR document, I think this is a name, I think this is a PO number, I think this is an address," Neelakantam said. It then checks those candidate entities against a canonical source a business has designated as reliable, such as a Salesforce record, and links matches into a knowledge graph. Where two systems disagree, both versions stay in the graph, tagged as related, rather than one getting silently discarded.

That matters more as agents start reasoning across systems instead of staying inside one application's walls. Neelakantam framed the shift as less about tearing down applications than about what sits above them: an agentic pricing engine, for example, isn't really an application in the traditional sense. It pulls from several systems, gets trained on success criteria, and plugs back in through APIs. For that to work reliably, the data underneath has to carry more context than a database column ever did on its own. Vocabulary is part of that problem. "All words are not created equal," he said. The word "inventory" means something different to a finance team than to a shipping team, and grounding that word in the actual documents and systems where it shows up, rather than treating it as plain English, is what lets a model tell the difference.

None of this means foundational models get it right by default once the data's connected. Neelakantam pointed to case-deflection systems as a concrete failure mode: an agent auto-closing a support ticket as resolved when the customer is still angry. "Nobody likes having a case auto-closed on them," he said. Fixing that kind of error isn't a matter of a new rule or a line of code, in his view. It takes the same discipline as traditional machine learning: logging inputs and outputs, running feedback loops, tracking precision and recall, and in some cases fine-tuning on the right training data. Prompt tuning alone doesn't get a system to production-grade accuracy.

The audit trail follows the same logic. As agents take on more autonomous action, Neelakantam said the industry is only beginning to build out what detailed logging of agentic workflows actually looks like — capturing intent, reasoning, and outcome at every step, not just the final answer. For regulated industries already running Data Intelligence in production, that's not optional. It's the same instinct driving the rest of this approach: don't hide the parts of the system that are hard to explain. Log them.

Neelakantam was candid that this transition carries real risk. Coding has gotten more accessible faster than the engineering discipline needed to deploy AI systems safely has caught up, in his view — a "wild west" period where people can build and ship agentic systems without necessarily understanding the failure modes underneath them. His read isn't that this slows down. It's that developers need to get comfortable fast with logging, testing, and documentation as the basic hygiene of handing real responsibility to a system that reasons on its own.

For developers building on top of any AI-readiness platform, not just Everpure's, the takeaway is less about the specific product and more about the posture: treat contradictory data as a signal, not a bug to be squashed before a model ever sees it. The version of your data with all its disagreements intact is more honest than the version an engineer smoothed over, and it's the version that will keep an agent from being confidently wrong.

Only 2% of companies are ready to scale AI securely—here's what the other 98% are missing.

Mon, 14 Jul 2025 11:00:34 +0000

The AI Readiness Reality Check: Why 98% of Organizations Aren't Ready to Scale

New research from F5 Networks delivers a sobering reality check for the enterprise AI landscape: while 96% of organizations are deploying AI models, only 2% qualify as "highly ready" to scale their AI initiatives securely. The findings, based on surveys of 650 global IT leaders and 150 AI strategists, reveal a massive gap between AI adoption and operational readiness that has significant implications for development teams.

The State of AI Infrastructure Today

According to Lori MacVittie, F5's Distinguished Engineer who led the research, the problem isn't lack of adoption—it's the fragmented approach organizations are taking. "The average organization uses three AI models, and this correlates with multi-environment deployments," MacVittie explains. "But most haven't solved the complexity and inconsistencies that come with that."

The data shows 25% of applications on average now use AI, but there's a stark divide in how organizations approach implementation. Highly ready organizations demonstrate "portfolio-wide saturation" while low-readiness organizations typically use AI in fewer than 25% of their applications, usually in isolated or experimental settings.

What sets the 2% of highly ready organizations apart? MacVittie points to several key factors: superior data practices, experience with diverse application types, and mature infrastructure management. "They understand how to formally manage data, how to label it correctly, and then secure it," she notes. "Larger organizations that have been doing this forever understand how to label data and use it and mine it already. So AI is just another thing they're going to mine, label, and secure."

The Security Paradox

Perhaps the most striking finding is a security paradox: 71% of organizations use AI to boost their security operations, but only 31% deploy AI-specific cybersecurity protections. This disconnect creates significant risks for development teams.

"Most security solutions have always had some form of AI in them already," MacVittie explains. "But the lack of AI-specific protections is concerning. You've got to defend against AI attackers, and they're not putting the types of AI protections that they need to identify, to watch, to do deeper inspection, and then be able to actually stop it."

The challenge is that AI attackers behave differently than traditional bots. "A normal bot will just do the URI as it can see, it follows expected paths. But an AI can take different paths. It can actually read the page and make a different choice. How do you distinguish between me, who's fumbling around on the website, and that AI?"

Data Labeling: The Foundation Missing

Only 24% of organizations practice formal data labeling—a critical gap that MacVittie identifies as essential for both transparency and preventing adversarial attacks. For engineering teams just starting to implement this, she recommends beginning with operational data labeling.

"Simple things like: this is an AI request from software bot, AI from human, part of a bigger process," MacVittie suggests. "Which process is this part of, the order process or the query process? Because these are two different security policies, I have to apply."

The research found that organizations taking a continuous approach to data labeling significantly outperform those using ad-hoc methods. "This kind of data labels are going to help security so much if you get it nailed down early," MacVittie emphasizes.

Cross-Cloud Complexity

As organizations deploy AI across multiple environments, governance gaps emerge. MacVittie provides a concrete example: "When you've got models in different places—maybe you're using OpenAI as a service and on-premise you might have an open source model like Mistral—you've got two different locations, two different models, and you've got two different security policies. So now you have inconsistency in security."

This fragmentation extends beyond just security policies to include access controls, data handling procedures, and basic AI governance—creating a complex web of inconsistencies that development teams must navigate.

The Observability Gap

One of the most critical missing pieces in AI infrastructure is what MacVittie calls "semantic observability." Current monitoring approaches are inadequate for AI systems.

"We monitor for up, down, fast, slow, somebody attacked, somebody kept making this request. But nowhere today do we really log things like what was the request. The prompt is now the context that we need to understand," she explains. "We need to be able to see that level and deeper—how did it answer and why."

This represents a fundamental shift for infrastructure teams. Traditional observability focused on system metrics, but AI requires understanding the semantic content of interactions between systems.

Technical Debt and Fragmentation

For the 77% of organizations classified as "moderately ready," the primary barrier to advancement is technical debt and infrastructure fragmentation. "Their infrastructure is fragmented," MacVittie notes. "We've got 35 different delivery and security things. We've got five WAFs. We've got three different kinds of load balancers. In order to scale, enterprises need standardization."

This fragmentation becomes particularly problematic as organizations try to implement consistent AI governance across diverse toolsets and environments.

Rethinking Application Architecture

MacVittie argues for a fundamental shift in how development teams think about AI systems. "We need to really embrace that these are all applications. Every single thing is an application. You're doing application to application," she explains.

This philosophical change has practical implications. Traditional client-server models break down in AI environments, where context flows bidirectionally and systems can switch roles dynamically. "These connections are going to go all over the place, so we've got to come to an understanding that's not how communication works anymore."

The Context Challenge

Perhaps the most significant architectural challenge developers face is managing context as state. "State is not stateless in AI systems. They carry the state with them," MacVittie explains. "As you're building an application, you have to be aware that as you're exchanging messages with other systems, especially other AI, that context is your state, and it might be corrupted. It might get too big."

This represents a fundamental shift from traditional stateless web architectures. Context in AI systems isn't just key-value pairs—it's large, complex text that must be actively managed throughout the application lifecycle.

The Path Forward

For development teams looking to improve their AI readiness, MacVittie recommends starting with the fundamentals: implement continuous data labeling processes, standardize infrastructure where possible, and begin building semantic observability capabilities.

"If you don't have a good observability practice right now, and most people don't because it was an afterthought, you need to lay it down now so that you're ready to be able to do this deeper logging and monitoring of what's going on as all this AI is talking to each other."

The window for addressing these challenges is narrowing. As AI adoption accelerates and agentic systems become more common, organizations that haven't built proper foundations will find themselves increasingly unable to compete. The 2% who are highly ready today aren't just ahead in deployment—they're building the infrastructure that will define the next generation of AI-powered applications.