Want to speak? Submit your talk and join our line up of speakers!
Community
Community
Overview
The story and values that drive us
Ambassadors
Become a Platform Engineering Ambassador
Events
Check out upcoming events near you
Reports
Check out the #1 source of industry stats
Jobs
Find your next  platform engineering role
Join Community
Join and contribute
Vendor opportunities
Certifications
Introduction to Platform Engineering
Platform Engineering Certified Practitioner
Platform Engineering Certified Architect
Agent infrastructure for Platform Engineers
new
Agentic Development Platforms
new
...and many more. Check out Platform Engineering University
Get Certified
For organizations
FOR ENTERPRISE TEAMS
Training & advisory
Home
Services
Results
Resources
FOR Partners
Service Provider
Training Reseller
Certified Provider Directory
BlogLandscape
Get certified
Join community
Join community
Get certified
All events
The future of AI-native is open source
Virtual
In-person
The future of AI-native is open source
Nov 5, 2026
7:00 pm
CEST
CET
-
45 minutes
Join Vultr for a session on why NVIDIA and AMD are opening the software above their silicon, and what that hands platform teams. Learn how to compare open AI stacks, test an opinionated stack against five criteria, and decide what to build, what to source, and what your team still owns.
Register
Watch recording
Speaker
Kevin Cochrane
CMO @ Vultr
Speaker
Speaker
Speaker

AI adoption is nearly universal, but real returns are not. In this conversation, Sam Barlien and Kevin Cochrane break down why open source AI infrastructure gives platform teams the control, safety, and flexibility to turn AI experiments into enterprise value - and why that responsibility lands squarely on your team.

Main insights

  • Nearly every organization uses AI, but only 8% report transformational ROI. Safety, governance, and identity are the differentiators, not raw model access.
  • Open source hardens AI infrastructure because community scrutiny finds vulnerabilities faster and gets them fixed faster.
  • Lightweight, post-trained open models support data sovereignty, predictable costs, and distributed deployment.
  • Platform teams must curate an opinionated, composable stack so developers build instead of assemble.

Sam Barlien is the co-author of the State of AI in Platform Engineering report and a long-time voice in the Platform Engineering Community. Kevin Cochrane is CMO at Vultr, the world's largest privately held independent cloud, with 34 global data center regions and a full-stack AI infrastructure offering built on NVIDIA and AMD GPUs. Together, they published The future of AI native is open source, and this session digs into its findings.

You can watch the full discussion here if you missed it.

Why ROI hinges on trust, not tooling

The State of AI in Platform Engineering research surfaces a striking gap: almost every team surveyed uses AI, yet only 8% see transformational ROI. What separates that 8%? According to Sam, it comes down to platform readiness, not model selection.

"All of them are thinking about sandboxing. They're thinking about governance. They're thinking about how they use governance to enable those agents," Sam explains.

Adding agents to your value stream is not enough on its own. The research shows agents alone make little difference "because you can't really trust them to go off and do the things you want." The value arrives when you layer in identity management and governance so agents can act safely within defined boundaries.

The same pattern shows up in what blocks AI at scale. Cost is not the top blocker. Identity, governance, security, safety, and trust rank higher - and those are fundamentally platform readiness problems. That is familiar territory for platform engineers.

How open source hardens agentic AI

Kevin argues that open source has always served two purposes: advancing enterprise infrastructure and securing it. The mechanism is community scale.

"When you have open source and you're bringing together the best and brightest minds on the planet to work together, that's how you're going to find your day zero bugs faster. That's how you're going to get them fixed faster. That's how you're going to harden the ecosystem."

That same energy is now turning toward agentic AI - AI systems that take autonomous actions within workflows. At the inaugural AgentCon and MCPCon events in Amsterdam and San Francisco, Kevin notes that half the conversation centered on safety, trust, and proper sandboxes, so teams can rely on more than guardrails alone.

Kevin also describes a proposal from frontier model providers - worth noting as a directional forecast rather than an established practice - where every frontier model would be made available to the wider community for external testing. Internal red teams, however skilled, carry inherent bias. Opening models to outside testers brings what Kevin calls "that hive mind mentality" that surfaces attack vectors internal teams miss.

The practical implication for your team: treat the agent sandbox as a first-class design concern, not an afterthought. Expect the open source ecosystem to supply much of the hardening you need, but you must architect for it deliberately.

Lightweight models, sovereignty, and the cost question

The future of enterprise agents is not simply API calls to a frontier provider. Kevin walks through why autonomous agents embedded in real workflows - such as optimizing logistics routing across a global supply chain - will often run on lighter-weight open models that you post-train on your own data:

  • Size: You are not going to deploy a 50,000 GPU cluster to run one workflow model.
  • Distribution: A global supply chain means running the model across multiple locations.
  • Data privacy: Post-training on historical shipment behavior, revenue, and margins uses data you may not want to send to a third-party provider.
  • Context: The agent also needs real-time connections to internal sources, such as inventory levels across different warehouses.

Data sovereignty strengthens the case further. Kevin gives the example of a UK financial institution training a credit assessment model on its historical loan data. That data must stay in the UK to comply with local regulation - a constraint that rules out many hosted model options.

Sam, based in Europe, adds a regional perspective. Europe lacks its own dominant models and clouds, so sovereignty must come from choice: "I have the ability to switch between a host of different self-hosted models or I have the freedom to switch between whomever based on whichever data centers I need." He calls this sovereignty through choice, and describes it as fundamentally an open source journey.

On the question every platform team eventually faces - "Is open source cheaper?" - both speakers say cost is only part of the answer. Consuming third-party tokens is essentially unbounded spend. Owning your reasoning models makes demand forecastable and capacity plannable, which matters to a CFO. A CSO cares for different reasons: data stays under your control. And models trained on your internal data make better decisions for your specific business context.

Sam points to Revolut's reported 90% cost reduction as an example, but notes the headline hides other gains: faster coding speed, easier work organization, and stronger data control. His practical tip for platform engineers: your role often includes translation. A CEO, a CFO, and a practitioner each need the same story told in different terms.

Building a composable, opinionated stack

Open source AI infrastructure is more than models. Kevin lists the surrounding components: FastMCP and the Model Context Protocol (MCP), serving and orchestration frameworks, and observability tooling such as OpenTelemetry applied to agents. Sam adds that open source tools like OpenClaw and NVIDIA's NemoClaw extend the ecosystem beyond model weights alone.

The risk is overload. Sam compares the open source AI landscape to the CNCF landscape, but accelerated. Kevin's answer is to assemble the pieces once, as a platform team, and distribute them as an internal distribution. An opinionated AI stack marshals best-of-breed open source projects and models into a curated starting point that downstream developers share - already tested, already instrumented for observability, traceability, and security.

Kevin describes the layering this way:

  1. Expose existing business functions as microservices.
  2. Wrap those services with MCP connectors so agents can call them without a human operating an interface.
  3. Package post-trained reasoning models as additional services - for example, a model trained on your ticket history for customer support triage.
  4. Curate the surrounding stack, including observability, test, and production infrastructure, as one starting point for developers.

"Exposing every single developer to saying, 'Hey, go figure everything out from scratch,' go to Hugging Face... that is way too much to ask any single human being to do," Kevin says.

The stack must stay composable. Kevin's position is that everything must be a fungible, ephemeral resource - models, serving frameworks, orchestration, and even the cloud beneath them. You can curate an opinionated stack, but you must be able to swap any element. That portability also keeps every layer of the ecosystem focused on being genuinely best-in-class.

Platform as a product still applies here. The way platform teams once enabled developers on Kubernetes distributions like OpenShift is exactly the model for AI tooling now. As Kevin puts it: "Yeah, tons of net new innovation, but does it wind up just being an extension of what we already know and do? And I believe firmly the answer is yes, 100%."

Where do you start? Kevin recommends focusing on internal use cases tied to developer productivity. Developers still deploy into your existing security and operations policies, so pre-build the agents that handle observability, security checks, and ongoing monitoring and remediation. Then make that stack available so developers can build without wiring up dashboards or failure-response agents themselves.

​

If you enjoyed this, find here more great insights and events from our Platform Engineering Community.

If you want to dive deeper, explore our instructor-led Platform Engineering Certified Professional course and connect with peers from large-scale enterprises who are driving platform engineering initiatives.

​

Key takeaways

  • Invest in trust before scale. Only 8% of teams see transformational AI ROI, and the leaders prioritize sandboxing, identity, and governance. Treat agent safety as core platform work, not an add-on. Without it, agents cannot be trusted to act autonomously in your value stream.
  • Evaluate open models for fit, not just price. Lightweight, post-trained open models support data sovereignty, predictable capacity planning, and distributed deployment. Translate the benefits into the language your CFO, CSO, and developers each care about - cost predictability, data control, and decision quality.
  • Curate one opinionated, composable stack. Package models, MCP connectors, OpenTelemetry observability, and security checks so developers build instead of assemble. Keep every layer swappable to protect against lock-in and maintain ecosystem leverage.
  • Apply what you already know. Platform as a product, golden paths, and developer enablement all carry over to AI. Start with internal developer productivity use cases, where your existing security and operations policies already apply, and expand from there.
This event is exclusive. Reserve your spot now.
Register now
Watch recording
Join our Slack

Join the conversation to stay on top of trends and opportunities in the platform engineering community.

Join Slack
Sitemap
HomeAboutAmbassadorsCertificationsEventsJobs
Resources
BlogBook: Thinking in PlatformsPlatformConCertified provider directoryWhat is platform engineering?Platform toolingVendor opportunities
Join US
Youtube
LinkedIn
Platform Weekly
Twitter
House of Kube
Weave Intelligence

Subscribe to Platform Weekly

Platform engineering deep dives and DevOps trends, delivered to your inbox crunchy, every week.

© 2026 Platform Engineering. All rights reserved.
Privacy Policy
Privacy PolicyTerms of ServiceCookies Settings
Supported by
Register now