Broadcom 观点：AI 暴露平台工程旧架构，需转向 Platform Engineering 2.0

partner content

The debate about whether enterprises need an internal platform is over. Google’s 2025 DORA research found that 90 percent of organizations now report using an internal platform, and 76 percent have established dedicated platform teams. The question for IT leaders is no longer “should we build a platform?” It’s “will the platform we’ve built survive what’s coming?”

The honest answer for most IT organizations is no — not without work. The platform you stood up over the past few years was designed for human developers shipping containerized apps at a human pace. That world is gone. Many forces have piled onto platforms in the last two years, and together they are exposing the cracks.

At the base of every crack is infrastructure — the developer-centric substrate that was never designed to provision GPUs on demand, govern AI agents, or enforce cost control at the moment of provisioning.

Start with the obvious one. Most developers are now using AI coding assistants. AI-assisted development has become mainstream, dramatically increasing the volume of code that organizations can generate and review. The bottleneck has moved. It’s no longer about writing code — it’s about delivering it. Your pipelines were never sized for that throughput, and the developer’s job has quietly changed from author to reviewer and orchestrator of machine-generated work.

Then there’s a brand-new kind of user knocking on the door: the AI agent. This is the non-human persona platform teams have increasingly had to design for over the last decade. Agents need authentication, token usage regulation and management at an organization and user level, GPU allocation, MCP compatibility, scoped permissions, non-human identity, audit logging, and hard guardrails on what they’re allowed to do. Most platforms have no native answer for any of it.

Cost is the third pressure. As per Broadcom’s Private Cloud Outlook 2026 study - 97 percent of IT leaders believe some of their public cloud spend is wasted, and 52 percent estimate that waste exceeds 25 percent of their total public cloud budget. AI infrastructure makes it dramatically worse. GPU instances, inference endpoints, and training jobs dwarf traditional spend. On top sits the token cost of every prompt and retry — a category most cost-reporting tools simply can’t see. Retrospective FinOps, the monthly review and quarterly cleanup sprint, can’t catch a misconfigured AI workload that burns through budget overnight.

Privacy, security and sovereignty round out the list. AI introduces new vast attack surfaces — shadow AI sprawl, prompt injection, model poisoning, inference data leaks — that no SAST or DAST scan in your pipeline was built to detect. Meanwhile the EU AI Act, US executive orders, and data residency rules keep adding compliance requirements.

Platform Engineering 2.0 Evolution, not a Rebuild

Here’s the part worth hearing clearly: none of this means tearing up what you’ve built. The foundations — Platform as Product, golden paths, shift-left security, self-service Internal Developer Platforms (IDPs) — still hold. The discipline doesn’t reset at each phase. It evolves. What’s described as Platform Engineering 2.0 is an extension of those foundations across five pillars, not a do-over.

The first pillar is an AI-native platform. Your IDP evolves into what’s increasingly being called an Agentic Development Platform (ADP), treating AI workloads as first-class citizens and AI agents as first-class users. Same operating model, expanded substrate.

Second, is a multi-persona experience. Most platforms today serve application developers well and treat everyone else as a second-class citizen. That leaves significant value on the table unrealized. Security teams, data scientists, ML engineers, FinOps analysts, business leaders, and agents all need the platform, each with their own interface sitting on shared APIs.

Third, is embedded FinOps. The shift is from rear-view-mirror reporting to cost decisions made at provisioning time. Every developer becomes a FinOps practitioner by default — not through training, but because the platform surfaces cost at the moment of action.

Fourth, security shifts down. Shift-left moved security earlier in the pipeline but dumped the load onto developers, and most vulnerabilities still surface in production. Shifting down means embedding security into the platform and runtime layers themselves, so it’s invisible to developers and immutable by design.

Fifth, composable by design. The future isn’t build versus buy — it’s compose. API-first, swappable building blocks let you replace one CI/CD tool or observability stack without cascading changes. The CNCF projects ecosystem has grown from about 50 projects in 2018 to more than 200 today; rigid platforms can’t keep pace with that choice surface.

Where to start: AI native platform readiness

Platform Engineering 2.0 begins with modernizing that infrastructure foundation. Don’t try to do all five at once. Audit your platform against the ceilings and start where the friction is worst — that’s where you’ll have executive sponsorship and a clear ROI story. For most organizations, the 12-month milestone that matters most is AI-native readiness: GPU workload support; agents; non-human identity; token economics; and MCP gateway exposure.

Start by auditing your current IDP for three gaps: GPU/accelerator provisioning; non-human identity management; and real-time cost attribution at provisioning time. If all three are missing, your platform is already behind.

The teams that built golden paths for developer autonomy are now being handed the keys to enterprise-wide agentic autonomy. That’s the largest mandate the discipline has ever held. The window is open and the foundations are already in place. What’s left is to extend them — before the platform becomes the bottleneck it was built to remove.

Contributed by Broadcom

When you just need to build a tiny desktop app: TinyJS

October 13, 2026

Valen creator drives 'Golden Spike' to connect new languages with Rust

September 26, 2026

Microsoft agentically ports Copilot runtime to Rust for $120K

September 19, 2026

Apple iPhone Duo makes developers think in folds

September 16, 2026

JavaScript installer pnpm recast in Rust because ECMAScript can't keep up

September 5, 2026

Kubernetes cleans house, bins legacy kube-dns, IPVS, and cgroup v1

August 27, 2026

How Cursor beat Git's scalability shortcomings

August 24, 2026

Go updates may delight diehard gophers but displease AI overlords

August 21, 2026

SvelteKit 3 puts heat on Next.js with radical approach to RPCs

August 20, 2026

Almost nobody pays attention to web standards anymore

August 18, 2026

ChainDrop worm crawls into npm supply chain, evades standard defenses

August 16, 2026

Node.js creator liberates Durable Objects from Cloudflare

August 13, 2026

AI titans to tidy agent frontier with plugin prescription

August 8, 2026

Platform Engineering 2.0: your platform was built for a different era. AI just exposed it

August 6, 2026

Next.js 16.3 aims to reduce dreaded FATAL ERROR messages

August 5, 2026

Dev who gave HashiCorp its name returns with a faster terminal multiplexer

August 1, 2026

How AI drove Shopify back to clean code

Turns out, agents just want the same things as humans: easily-readable code, explicit contracts, and helpful feedback

July 26, 2026

Zig creator calls Bun's Claude Rust rewrite 'unreviewed slop'

July 15, 2026

HTTP gets a QUERY method so complex searches can stop pretending to be POST

July 14, 2026

Speedier type checks in TypeScript 7.0 as first stable Go release ships

July 10, 2026

GitHub cuts short offer to burn repos on CD after mockery ensues

A purported jab of Sony's physical media phase-out blows up on GitHub itself

July 7, 2026

.NET's long term support is not long term enough, dev complains

June 30, 2026