In this exclusive interview with Swapnil Bhartiya at TFiR, Jeremiah Stone, Chief Technology Officer at SnapLogic, discusses how SnapLogic’s Agentic Integration Platform — anchored by the newly announced AI Gateway and Trusted Agent Identity — is solving the enterprise AI execution problem by turning AI agents from sandbox demonstrations into governed digital labor that operates inside real business systems.

What SnapLogic Does: Bridging Data and Application Integration for the Agentic Era

SnapLogic has spent years focused on reducing the human cost of integration across heterogeneous enterprise environments — spanning SaaS, on-premise, cloud, and hybrid deployments. Its platform spans both data integration (moving and normalizing data across systems) and application integration (taking action within enterprise applications), a combination that positions it uniquely for the demands of agentic AI workflows.

Q: Tell us about SnapLogic — what does the company do and what makes it different?

Jeremiah Stone: “SnapLogic is a company that firmly believes in accelerating success in enterprise environments. We are fundamentally an integration, automation, and orchestration platform. We have focused on bridging and unifying what have typically been separate worlds. There are many people that focus on data integration — getting access to data within the enterprise. There are many that focus on application integration — applying action to enterprise systems. SnapLogic is unique in bridging these worlds. We can as easily interoperate and unify normalized data systems as well as we can take action in application systems. And that set us up to be really thoughtful, primed, and ready to focus on the rise of digital labor.”

Stone explained that the concept of digital labor — and the emerging practice of digital shoring — requires thinking through every element of what it means to get work done inside the enterprise: access control, workflow management, orchestration, and action in application systems. This complete span is what SnapLogic targets.

Q: What does “applied AI infrastructure” actually mean in practice?

Jeremiah Stone: “If you think about what applied AI infrastructure really means, you have to have ubiquitous access across all of the individual data systems. You need to be able to transform and normalize that data so agents can use it. You then need to be able to orchestrate within the governance and rules of the business and take action in the applications of the business itself. And so we focus across that entire span.”

Defining Digital Labor: What Agents Are Actually For in the Enterprise

Before discussing the infrastructure that enables digital labor, Stone grounded the conversation in what digital labor actually means in a business context — and why unit economics, not technology novelty, is the right frame for evaluating it.

Q: How do you define digital labor, and what kinds of enterprise tasks does it apply to?

Jeremiah Stone: “Fundamentally, the reason why agents are interesting in the business sense is unit economics. Unit economics means how much does it cost me and how much time does it take to get a unit of work done — a task. When we think about what is digital labor in the simplest sense of the term, it means that we’re able to automate work and have the machine do tasks or activities that were previously only possible with human intervention. In the context of the enterprise, much of digital labor is actually highly repeatable processes that have heterogeneous data — some of it transactional, like from a CRM system or an ERP, some of it unstructured: documents, purchase orders, invoices. The labor that could not be automated before language models was labor that had to take these types of information together, use contextual reasoning, and then take action in those systems. When we talk about digital labor in the context of the enterprise, we’re talking about high-volume, heterogeneous-data tasks that require contextual reasoning and action within enterprise systems.”

AI Gateway: A Governed Front Door for Agentic Systems

The AI Gateway is SnapLogic’s application of API management, workflow, and policy-based governance principles to the agentic web. It addresses the full lifecycle of agent interaction within the enterprise — from authentication and authorization through traffic management and observability — applied to MCP-connected tools and services.

Q: What is the AI Gateway and why do AI agents need it?

Jeremiah Stone: “When we talk about an AI gateway, we really mean the end-to-end application and usage of the enterprise environment. That means you have to be able to both make the enterprise surface — whether those are APIs, documents, file systems — available to the agents, but you also have to be able to manage the work as well. You have to be able to authenticate the agent coming in to make sure that it is able to take action within the systems. You have to authorize — authentication and authorization are different things. But then you also have to manage traffic across all of the AI interactions. At the moment, organizations are using MCP as a way to expose the different tools and services within the enterprise, but when an agent is hitting an MCP endpoint, that agent needs to be able to authenticate, set its scope and context, and then be able to be managed in a way that doesn’t blow up the underlying systems and create enormous latency or other problems. The SnapLogic AI Gateway takes principles from API management, from workflow, from policy-based management, and applies it in the context of agentic systems.”

Q: How does AI Gateway fit into the broader SnapLogic Agentic Integration Platform?

Jeremiah Stone: “For many years, we have focused and worked with our customers on driving the human cost of integration to zero — low-code, no-code, direct connectivity, drag-drop-configure in a visual interface. Making all enterprise systems in heterogeneous environments, whether SaaS or on-prem, available to be integrated. What we’re adding now is the ability to orchestrate AI workflows. Whether you’re using the SnapLogic AgentCreator — our capabilities to build agents within the SnapLogic platform — or you’re using a different agentic platform, we can now orchestrate workflow across heterogeneous enterprise environments and manage governance, security, and observability. This is an additional layer enhancing the capabilities SnapLogic has already provided, but built from the ground up for the agentic web. And we put it to work in our own business.”

Trusted Agent Identity: Propagating Access Control Across Enterprise Systems

Trusted Agent Identity solves the authorization propagation problem that has blocked most agentic POCs from reaching production. By working directly with enterprise identity providers (IDP), SnapLogic enables organizations to pass user context — including OAuth tokens — through the gateway and into every downstream system, creating consistent, auditable, policy-enforced access across all AI models and enterprise applications.

Q: Why do AI Gateway and Trusted Agent Identity belong together?

Jeremiah Stone: “The SnapLogic AI Gateway works seamlessly with the enterprise’s identity provider. When the user is authenticating — let’s say they’re using an agent in a conversational way — we can take the OAuth token from the user, pass that via the gateway, and then propagate that to the underlying systems. Because you’re using the identity management provider, you’re able to manage the session across systems. We have customers that find it really valuable to mix and match between Gemini, Claude, OpenAI, or even something like Mistral or a model from Hugging Face running in their own serving infrastructure. Because you’re now normalizing and standardizing your access control, it dramatically simplifies the architecture — it doesn’t complexify it. Quite the opposite. You can manage it in a well-governed way across the entire system.”

Addressing MCP’s Production Gaps

Stone called out a critical blind spot in emergent agentic web standards — including MCP (Model Context Protocol) and A2A — that SnapLogic is specifically working to address. These standards do not solve the enterprise governance, authentication, and orchestration requirements needed for production deployment in regulated industries.

Q: What are the gaps in current agentic web standards like MCP and A2A?

Jeremiah Stone: “The emergent standards have kind of ignored this area. If you think about agentic web standards like MCP or A2A — we’ve really focused on hardening these emergent standards so they can be put to work in production environments in industries like finance, healthcare, heavy industry, and manufacturing. Our most recent announcements are providing the ability to have a secure and governed front door in terms of an AI Gateway, but then the ability to manage access from that gateway down in and across all of the individual applications and systems within the enterprise itself.”

Customer Impact: From POC Theater to Production Reality

Stone shared concrete examples of how enterprises are applying SnapLogic’s capabilities to transform business operations — not as AI experiments, but as fundamental changes to unit economics and cycle times across financial services, HR services, discrete manufacturing, and biotech.

Q: What does this actually change for organizations that adopt these capabilities?

Jeremiah Stone: “One of our customers in the HR services realm is focusing not on software as a service, but service as software. They’re looking at the services they’ve delivered as a business over the last 20 or 30 years, finding the elements they can completely automate, and putting their domain knowledge and proprietary data to work in a way that completely changes the economics of their business. They’re redefining their entire business model to compete more effectively and to become an agentic company — working backwards from the business outcome, decomposing all of the processes and tasks associated with that outcome, and then automating those individual tasks.”

Jeremiah Stone: “We have customers in the finance domain where they are changing business processes for loan underwriting. They used to take 48 hours to go through the process of onboarding and application, doing the different credit checks, anti-money laundering checks. Now they can perform all the checks in 30 minutes. From 48-hour cycle time to 30-minute cycle time on key business processes — and completely automated. Not just offshored to different service centers. It’s a complete, disruptive, innovative leap ahead. We’re seeing it in discrete manufacturing, financial services, biotech, and business service delivery.”

Q: What is the common thread across all these transformations?

Jeremiah Stone: “The common thing across all of these is the ability to look at a business outcome, decompose the tasks associated with it, identify the heritage systems and the landscape that will support the agents, and then build and deploy the agents. It’s taking the best of historical enterprise technology and combining it with the agentic technologies from the AI labs or from the larger ecosystem, whether open source or commercial vendors.”

Breaking the POC-to-Production Wall

Stone was candid about why most AI proof of concepts fail to reach production — and specific about what AI Gateway and Trusted Agent Identity unlock to change that dynamic.

Q: Why do nearly half of generative AI POCs get abandoned, and what does SnapLogic do about it?

Jeremiah Stone: “Unfortunately, most proof of concepts are success theater. They frame the data upfront in a way that they can use it. It’s mostly used to learn how to use the generative AI technology itself. And then it demonstrates within a completely fake and constrained environment that the team was able to understand the business context, maybe. But the ability to put that in production — they haven’t even tackled any of the real problems. What AI Gateway allows these teams to do is take, as part of the proof of concept process, the deep responsibility to also prove business value by utilizing the governance frameworks and rules required for production deployment — and build that into the proof of concept. So you can knock down technical and business risk in the POC phase already.”

Q: What does the failure pattern actually look like in practice?

Jeremiah Stone: “What I see happen — the majority of proof of concepts that fail to move to production — is that after they prove the idea of the agentic model could work for the business context, they take on the rest of the iceberg instead of retiring that risk early and building it into the POC itself. AI Gateway allows these same teams to take a thin slice and retire the infrastructure needs as well. We’re seeing many teams that struggled in POCs — maybe spent a year and a half using only open source tooling, hand-building all of these things — only to find out they’re going to have to hand-build their authentication mechanisms, authorization mechanisms, orchestration, and policy management. They’re able to pivot very quickly into using SnapLogic tooling and accelerate. Going from a year and a half of struggling to maybe six to eight weeks being able to pivot a project into success.”

Q: What is your guidance to teams launching agentic programs today?

Jeremiah Stone: “My guidance and recommendations to teams launching into these domains is to identify the governance and larger security management burdens they’ll have to bear, and start to at least test and burn those down early in the process. We love it when teams start from the beginning, retiring technical and business risk as part of the proof of concept — really being a proof of value.”

Multi-Model, Multi-System Complexity: How AI Gateway Handles 1,000+ Connectors

SnapLogic supports over 1,000 enterprise connectors and is model-agnostic across OpenAI, Azure OpenAI, Google Gemini, Amazon Bedrock, Mistral, and models running on Hugging Face. Stone explained how AI Gateway manages this complexity by standardizing at the identity provider level.

Q: How does the AI Gateway handle multi-model, multi-system complexity without creating new governance nightmares?

Jeremiah Stone: “Fundamentally, we thrive in the spaces between the systems — that’s our goal. Our job is to make all of these complex, diverse system environments, whether it’s the network architecture, the deployment models, the different systems, or even language models, just work seamlessly together. What we’re doing is making the AI Gateway work seamlessly with the enterprise’s identity provider. Because you’re now normalizing and standardizing your access control, it dramatically simplifies the architecture. We have many customers that are highly heterogeneous in their model usage. We partner closely with the hyperscalers — AWS on Bedrock, working with the AI labs — to make sure we can even out all of these differences between these systems.”

Jean-Paul: SnapLogic’s Internal MCP-First AI Business Partner

Jean-Paul is SnapLogic’s general-purpose internal AI agent — built on an MCP-first architecture and deployed across the company’s own systems. The adoption numbers Stone shared make it one of the most compelling internal AI deployment stories in the enterprise software space.

Q: What is Jean-Paul and why did SnapLogic build and open it to customers?

Jeremiah Stone: “Jean-Paul is our general-purpose business partner for every employee at SnapLogic. It’s an AI agent — but really a general-purpose business partner that can work across all of our internal systems and meet the employee where they are. Whether it’s email, Slack, Teams, scheduled workflows — it’s really a teammate and a business partner for us. Because we are a software ISV, that means we need to focus on the entire business lifecycle: prospecting for new customers, going through a sales process, supporting a renewal or an expansion with an existing customer, understanding our internal platform operations — stitching across all of our individual systems, whether that’s our ERP system, our sales system, or our unstructured documents. Jean-Paul has helped normalize that across the system for us and also generate documents — whether a customer-facing presentation or a document for a board meeting.”

Q: What adoption metrics did Jean-Paul achieve internally?

Jeremiah Stone: “In the first 30 days of internal production use with zero training, we were able to scale up to 100 users across 16 departments. We’re talking hundreds of thousands of dollars in productivity value in the first 30 days and thousands of hours of work saved. We found this so valuable and impactful for our business — we’re an applied AI company, but this is a really valuable application. We’re making it available as an experiment, as a gift from our learning, because we’ve been working in this domain for seven or eight years now. We have numerous early pilots on Jean-Paul already and it’s going really well.”