This year’s PlatformCon Live Days in London made one thing clear: AI is dominating the architectural landscape, and it isn’t only transforming software engineering. It’s raising the stakes for platform engineering.
Across talks from practitioners, a consistent theme emerged: the organizations that succeed with AI won’t be those with the flashiest models, but those with the most resilient, adaptive platforms.
Between packed presentation tracks and useful peer discussions in the “Platform Therapy Room,” there was an immense amount to process. Here we summarize our team’s top four themes, along with our favorite highlights from the event.
Key Themes
1. Managing the Non-Deterministic Beast: Controls & Responsibility
A major talking point was the fundamental architectural tension between AI and platform stability: AI systems are probabilistic, but platforms must remain deterministic.
The Four Levels of AI Development: To scale, we have to accept the non-deterministic nature of AI, establish clear controls, learn to trust those controls, and finally “unleash the beast” and let it fly.
Lane Assist over Guardrails: As Gregor Hohpe noted during the event, “No one drives their car by bumping along the guardrails. You need lane assist as well as guardrails.” Quantifying risk and maintaining AI responsibility in a regulated environment is a serious hurdle when the same prompt can yield different results. Platforms must actively guide users toward success rather than just enforcing restrictions.
2. From IDPs to ADPs: Scaling for the Agentic Enterprise
Internal Developer Platforms (IDPs) provide an excellent baseline, but the future belongs to the Adaptive / Agentic Developer Platform (ADP).
The “Dark Factory” and New Users: Platforms are expanding to support a hybrid mix of human developers, copilots, and autonomous agents, approaching Patrick Debois’ concept of a “dark factory.”
Democratizing the Platform: We are starting to see requirements come from far beyond traditional development teams. Everyone across the enterprise wants to request agent functionality, often without the technical background to build it themselves. The next major hurdle is making these platforms intuitive and accessible for non-technical users.
3. Sovereign Infrastructure vs. Escalating Cloud Costs
With the rise of large GenAI workloads, cloud costs are climbing rapidly, forcing teams to constantly trade off between cost and infrastructure sovereignty.
The On-Prem Hardware Resurgence: In a surprising shift to avoid extreme vendor lock-in and sky-high cloud bills, many companies are actively evaluating, or have already committed to, buying their own hardware. As our team noted from the floor: “There’s always a vendor to take your money and tell you your bad idea is a good idea.”
Platforms That Float: To protect sovereignty and let companies move workloads dynamically, Infrastructure as Code (IaC) has become vital. Successful platforms must “float” above commodity cloud services, focusing on organization-specific workflows and policies rather than reinventing the wheel.
4. High-Throughput Code and the Collapse of the Traditional Lifecycle
When using AI agents, the measure of success shifts to throughput over speed. But producing code in seconds exposes serious downstream operational bottlenecks.
Is Git Reaching its Limit? One of the most surprising realizations of the event is that we may have to completely reconsider traditional workflows like Git for agentic code management. It simply won’t keep up with the hyper-scale throughput of autonomous development.
The Death of Patching: With AI accelerating both code generation and exploit discovery, the security panel observed that “patching is doomed.” To keep pace with the vulnerabilities found in that volume of generated code, organizations must heavily invest in automated testing, runtime security, and secure platform defaults.
Highlighted Sessions & Moments
Welcome Address: “AI Needs Platforms More Than Platforms Need AI” Kaspar von Grünberg & Luca Galante set the scene early by arguing that AI-based software engineering cannot succeed without effective platforms. An AI agent can propose architectures or spin up pull requests, but the platform must deterministically provide the human override mechanisms, sandboxed environments, governance, and rapid feedback loops.
The “Platform Therapy Room” Roundtables Arguably our team’s favorite part of the live event, these roundtables sparked incredibly rich debates. Discussions dove straight into the technical weeds of agentic development while safely tackling the moral questions of AI autonomy. It was a great forum to debate building custom solutions in-house versus using third-party libraries when navigating highly regulated environments.
“GenAI Amplifies Your Dysfunctions” Gregor Hohpe delivered what might be the quote of the conference: “GenAI does not make your problems go away. It amplifies your dysfunctions.” He reminded everyone that if developers can generate code in seconds but still wait days for environment provisioning, security reviews, or governance sign-off, the platform friction becomes painfully obvious.
Scaling AI Capabilities at Booking.com Mansi Mittal & Bruno Passos shared practical lessons from scaling AI at Booking.com. They proved that successful adoption isn’t about model selection; it’s an organizational capability. Real value comes from putting deliberate education, enablement, measurement, and community feedback loops around the platform.
Looking Ahead
PlatformCon 2026 proved that while many answers about a fully autonomous, agentic future remain unclear, platform engineering is the critical connecting layer that will pave the way. As software generation accelerates into overdrive, the platform becomes the mechanism to operationalize AI safely, securely, and cost-effectively.
We are excited to tackle these challenges head-on within our own architectures, and we look forward to seeing how these trends mature by next year. If your platform team is wrestling with the same questions, we are happy to compare notes.
