limitedDistribution · Industry Research
Interoperability in Supply Chains
AI infrastructure for laboratories is becoming less about a single model or tool and more about the systems that let research environments connect, share data,.
AI infrastructure for laboratories is becoming less about a single model or tool and more about the systems that let research environments connect, share data, and operate reliably. According to Lab Manager, laboratories moving toward more connected and potentially autonomous workflows need to treat data architecture and interoperability as foundational management considerations. Lab Manager also reports that advanced research infrastructure increasingly depends on coordination across physical lab systems, digital resources, data practices, and specialized technical expertise. That shift aligns with how agentic AI systems are being structured. arxiv.org describes protocols as a way to externalize interaction structure in LLM agents, which matters because structured interactions can help define how AI components coordinate with tools, systems, or users. For buyers and lab leaders, the practical implication is that AI readiness should be evaluated alongside interoperability, data practices, and technical support capacity—not just model performance. Communications infrastructure is part of the same resilience picture. AOL.com reports that Avaya Aura 10.3 continues JITC certification and meets cybersecurity and interoperability requirements for U.S. federal and military communications environments, underscoring that secure, interoperable systems remain central where operational reliability is required.
Key Takeaways
- The timing matters because AI-enabled research is moving from isolated, well-funded environments toward shared regional infrastructure.
- Trend 1: Shared AI infrastructure is becoming part of lab planning Laboratories are beginning to treat advanced computing access less as a standalone IT purchase and more as a shared infrastructure strategy.
- The second trend is the shift from “better model” thinking to “better runtime” design.
- Trend 3: Modernization is shifting from rip-and-replace to protected evolution For large enterprises and government agencies, the modernization path is increasingly being framed around continuity: keep proven communications infrastructure in place, strengthen its security posture, and prepare it for future capabilities without forcing a premature platform replacement.
- The operational impact is that AI readiness becomes less about a single model or compute purchase and more about the laboratory’s surrounding systems: data practices, repeatable procedures, security governance, and tool coordination.
The timing matters because AI-enabled research is moving from isolated, well-funded environments toward shared regional infrastructure. According to Lab Manager, the US National Science Foundation has launched a $100 million program to expand researchers’ access to computing resources, technical expertise, and training for AI-enabled scientific research. That shifts the practical question for many laboratories from whether they can afford to build advanced computing capacity on their own to whether they can connect their people, data, and workflows to emerging shared resources. The program also broadens the ecosystem around laboratory AI. Lab Manager reports that participating institutions will work with partners that can include state and local governments, industry, and philanthropic organizations to make computing, data, software, and other AI resources available to researchers, students, and educators. For labs, this creates a new access model: shared infrastructure may offer another route to advanced computing without requiring every research organization to develop the same capabilities independently. But this is not just a compute story. Lab Manager emphasizes that access to compute alone does not make a laboratory AI-ready; labs still need experimental data that are structured, contextualized, accessible, and suitable for analysis. That makes the current moment especially important for managers evaluating data systems, instrument connectivity, metadata practices, and interoperability. As organizations move toward more connected and potentially autonomous workflows, data architecture is becoming a foundational management concern, not a back-office IT issue. The opportunity is expanding, but so is the need to prepare data environments so they can actually use the AI infrastructure becoming available. One trend is that shared AI infrastructure is becoming part of operational planning. Laboratories are beginning to treat advanced computing access less as a standalone IT purchase and more as a shared infrastructure strategy. According to Lab Manager, the NSF’s regional AI infrastructure hub program is intended to broaden access across institutions of different sizes, including smaller institutions and community and technical colleges that may have less independent access to advanced computing infrastructure. That matters because many labs may need AI-ready compute, data systems, and technical expertise before they can justify building or maintaining those capabilities on their own. Lab Manager reports that shared infrastructure could provide laboratories with another route to advanced computing without requiring every research organization to develop the same capabilities independently. In practice, this shifts the planning question from “Should this lab own all the infrastructure?” to “Which capabilities must be internal, and which can be accessed through coordinated regional or institutional resources?” The trend also reaches beyond compute capacity. Lab Manager found that as laboratories move toward more connected and potentially autonomous workflows, data architecture and interoperability become foundational management considerations. That means AI infrastructure planning increasingly includes how instruments, data repositories, workflow software, and governance practices connect. The broader signal is that advanced research support now depends on coordination across physical laboratory systems, digital resources, data practices, and specialized technical expertise. For leaders, the emerging priority is not simply adding AI tools, but designing an operating model that can use shared infrastructure reliably, securely, and consistently across workflows. A second trend is the shift from “better model” thinking to “better runtime” design. According to arxiv.org, Zhou et al. describe LLM agents as increasingly being built by reorganizing the infrastructure around the model rather than changing the model weights themselves. In this view, reliability is not only a property of the base model; it is also a property of the environment that stores context, structures tool use, constrains execution, and makes behavior observable. Zhou et al. define this pattern as externalization: relocating cognitive burdens from model-internal computation into persistent, inspectable, and reusable external structures. That matters because many operational problems in agents are not solved simply by asking the model to “reason better.” They are addressed by giving the agent durable memory, reusable skills, standardized tool interfaces, constrained execution paths, behavior instrumentation, and explicit control logic. These elements turn one-off prompts into a governed operating environment. Protocols are a key part of this shift because they externalize interaction structure. Instead of leaving every exchange to ad hoc model behavior, protocols define how the agent should interact with tools, users, systems, and intermediate outputs. Harness engineering then coordinates memory, skills, and protocols into governed execution, making the agent less dependent on implicit behavior and more dependent on inspectable system design. The practical implication is that teams seeking reliability gains should look beyond model selection. Per arxiv.org, Zhou et al. argue that practical LLM agent reliability often improves by changing the model environment rather than the base model. For buyers and builders, that reframes evaluation around runtime architecture: what the agent can remember, how it invokes tools, how execution is constrained, and how behavior is monitored over time. A third trend is that modernization is shifting from rip-and-replace to protected evolution. For large enterprises and government agencies, the modernization path is increasingly being framed around continuity: keep proven communications infrastructure in place, strengthen its security posture, and prepare it for future capabilities without forcing a premature platform replacement. According to AOL.com, Avaya Aura 10.3 is intended to give these organizations a secure, supported path to modernize communications environments while preserving technology that is already working. That positioning matters most in environments where communications platforms are deeply embedded in operations, compliance processes, and mission workflows. Avaya Aura 10.3 continues JITC certification and meets cybersecurity and interoperability requirements for U.S. federal and military communications environments, making the release relevant to buyers that cannot separate modernization from certification and assurance requirements. The trend is also visible in the platform’s infrastructure updates. Avaya Aura 10.3 extends Trellix antivirus support to JITC-compliant environments, updates key third-party software components, upgrades to Red Hat Enterprise Linux 9.6, and supports VMware ESXi 9.x. Those changes point to a modernization model based on supported operating environments, security maintenance, and interoperability rather than wholesale disruption. Tony Lama, SVP and GM of Avaya Software, said the release lets customers modernize on their own terms while protecting their current ecosystem and enabling future capabilities. In practical terms, that makes Aura 10.3 part of a broader enterprise communications trend: modernization that reduces risk by respecting what is already deployed. For supply chain organizations, interoperability is most valuable when it turns external partner inputs into usable workflow data—not just when systems are technically connected. Stargo benchmarks show AI-assisted vendor document normalization cut supplier onboarding cycle time by 11.4 days on average, while one deployment processed 18,000 purchase-order attachments in its first 30 days without adding back-office headcount. The implication: interoperability investments should be judged on whether they reduce manual translation between suppliers, procurement systems, and approval workflows—not only on API availability or platform compatibility.
Operational Impact
The operational impact is that AI readiness becomes less about a single model or compute purchase and more about the laboratory’s surrounding systems: data practices, repeatable procedures, security governance, and tool coordination. According to Lab Manager, access to compute alone does not make a laboratory AI-ready; experimental data still need to be structured, contextualized, accessible, and suitable for analysis. That shifts day-to-day responsibility toward data stewardship: standardizing how experimental context is captured, making records retrievable for analysis, and reducing the amount of manual reconstruction needed before AI tools can be useful. For teams deploying LLM-based assistants or agents, the practical constraint is consistency. arxiv.org reports Zhou et al.’s findings that unaided LLMs face a continuity problem because of finite context windows and weak or absent session memory, a variance problem when long multi-step procedures are rederived instead of executed consistently, and a coordination problem when work with tools, services, and collaborators depends on free-form prompting alone. Operationally, this means labs should avoid treating prompts as informal know-how. Procedures, handoffs, tool permissions, and data access rules need to be externalized into controlled workflows where possible, so AI-enabled work can be repeated, audited, and improved. Cybersecurity also becomes part of the operating model rather than a separate back-office concern. Lab Manager reports that the NSF program puts cybersecurity within the infrastructure workforce it plans to support, and notes that lab managers increasingly have a role in cybersecurity governance and risk awareness even when dedicated IT teams retain technical control responsibilities. In practice, AI infrastructure expansion requires closer coordination among principal investigators, lab managers, data stewards, and IT teams so that access, risk awareness, and research productivity advance together.
What Buyers Should Evaluate
- Buyers should evaluate AI infrastructure and communications upgrades as operating models, not just product selections. The first screen is whether the offering covers the full stack needed for sustained use: compute hardware, data infrastructure, software, networking, storage, cloud services, and related AI systems. According to Lab Manager, regional AI infrastructure consortia are expected to secure these capabilities through institutions, governments, industry, philanthropy, or other sources, which makes supplier ecosystem strength and funding flexibility important evaluation criteria. The second screen is workforce readiness. Buyers should ask whether the vendor, integrator, or consortium can support systems administration, storage architecture, cybersecurity, network engineering, software engineering, training, and user support. Lab Manager also identifies data engineering and curation, research software engineering, model deployment, GPU programming, access control, and secure research-environment management as workforce priorities. That means buyers should assess not only platform specifications, but also who will operate, secure, tune, and train users on the environment over time. For agentic AI or LLM-enabled systems, buyers should look for reliability features in the surrounding environment. arxiv.org reports that persistent memory, reusable skills, standardized tool interfaces, constrained execution, behavior instrumentation, and explicit control logic can improve agent reliability, while evaluation, governance, and long-term co-evolution between models and external infrastructure remain open challenges. Procurement teams should therefore require evidence of evaluation methods, governance controls, observability, and upgrade paths rather than accepting model performance claims alone. For federal, military, or regulated communications environments, compliance and interoperability should be explicit gates. AOL.com reports that Avaya Aura 10.3 continues JITC certification and meets cybersecurity and interoperability requirements for U.S. federal and military communications environments. Buyers considering that type of deployment should also plan upgrades with the account representative or authorized partner, since supplier-directed upgrade planning is part of the recommended path.
Definitions
Externalization: In LLM-agent systems, externalization means moving cognitive work out of the model’s internal computation and into structures that are persistent, inspectable, and reusable. According to arxiv.org, Zhou et al. define externalization as relocating cognitive burdens from model-internal computation into external structures that can be retained, examined, and reused. Memory: Memory is an external structure that carries state across time for an LLM agent. arxiv.org reports that memory externalizes state across time, which means the agent can rely on stored context rather than only the immediate model computation. Skills: Skills are externalized procedural expertise. Per arxiv.org, skills externalize procedural expertise in LLM agents, giving the system reusable ways to perform tasks rather than forcing each procedure to be reconstructed from scratch. Protocols: Protocols are externalized interaction structures. arxiv.org found that protocols externalize the structure of interactions, helping define how an agent should coordinate steps, exchanges, or roles. Harness engineering: Harness engineering is the layer that coordinates memory, skills, and protocols into governed execution. In practical terms, it is the organizing approach that makes these externalized components work together under defined controls. AI infrastructure access: Lab Manager reports that participating institutions can work with state and local governments, industry, and philanthropic organizations to make computing, data, software, and other AI resources available to researchers, students, and educators. JITC-certified communications environment: AOL.com reports that Avaya Aura 10.3 continues JITC certification and meets cybersecurity and interoperability requirements for U.S. federal and military communications environments.
FAQ
FAQ Q: What does it mean for a lab or organization to be “AI-ready”? A: AI readiness is not just a matter of getting access to more compute. According to Lab Manager, laboratories also need experimental data that is structured, contextualized, accessible, and suitable for analysis. In practice, that means teams should look at data quality, metadata, system connectivity, and whether results can be reused by analytical tools before assuming AI infrastructure will deliver value. Q: Why are data architecture and interoperability showing up in AI infrastructure discussions? A: Lab Manager reports that as laboratories move toward more connected and potentially autonomous workflows, data architecture and interoperability become foundational management considerations. If instruments, databases, workflow systems, and AI tools cannot exchange information reliably, automation and AI-assisted analysis are harder to scale. Q: How do protocols relate to LLM agents? A: arxiv.org describes protocols as a way to externalize interaction structure in LLM agents. The core issue is coordination: arxiv.org reports that unaided LLMs face a coordination problem when interactions with tools, services, and collaborators depend only on free-form prompting. Protocols can help define how an agent should interact with external systems instead of leaving every step to ad hoc language instructions. Q: What should buyers evaluate when comparing AI-enabled workflow systems? A: Buyers should examine whether the system can work with structured and contextualized data, whether it supports interoperability across existing tools, and whether agent interactions are governed by clear protocols rather than informal prompts alone. These criteria matter because the available evidence points to data readiness, connectivity, and coordination as recurring constraints. Q: How does communications infrastructure fit into this topic? A: For some environments, especially government or defense settings, AI-enabled workflows may depend on secure and interoperable communications foundations. AOL.com reports that Avaya Aura 10.3 continues JITC certification and meets cybersecurity and interoperability requirements for U.S. federal and military communications environments. That example shows why security certification and interoperability can be procurement considerations alongside AI features themselves.
Stargo insight: Interoperability is a workflow outcome, not just a systems feature
For supply chain organizations, interoperability is most valuable when it turns external partner inputs into usable workflow data—not just when systems are technically connected. Stargo benchmarks show AI-assisted vendor document normalization cut supplier onboarding cycle time by 11.4 days on average, while one deployment processed 18,000 purchase-order attachments in its first 30 days without adding back-office headcount. The implication: interoperability investments should be judged on whether they reduce manual translation between suppliers, procurement systems, and approval workflows—not only on API availability or platform compatibility.
Related guides: Transport in Supply Chain: What Buyers Need to Know, Supply Chain Solutions in the AI Era.
See ROI in 12 weeks
Stargo users see measurable return and operational profitability gains in just 12 weeks, with non-disruptive implementation in 4 weeks or less.