Technology
November 15, 2026 1003 words

AI Developer Infrastructure: llm-d 1.0, AWS Emulator, Trajectory and the Tools Behind Agents

Hams AI Tech
Professional Content Team
Expert in AI Prompts & Professional Tools

Headlines focus on new models, but much of the progress in AI comes from the infrastructure that serves, tests and improves them. In October and November 2026, several important developer tools reached maturity: llm-d 1.0 for distributed model serving, a free local AWS Emulator that coding agents can test against, Trajectory's platform for post-training models on production data, and OpenAI models arriving on Google Cloud. If your organization builds AI tools, or pays someone who does, these releases affect cost, reliability and speed.

llm-d 1.0: Production-Grade Distributed Inference

In late October, the llm-d project released version 1.0. Governed by the Cloud Native Computing Foundation and founded by Red Hat, IBM, Google Cloud, CoreWeave and NVIDIA, llm-d is a hardened open-source core for distributed large language model inference. It is aimed at production deployments and context-heavy agentic workloads.

Why does that matter? Serving large models efficiently across many GPUs is difficult. Agentic workloads make it harder, because agents send long contexts repeatedly as they work through tasks. A mature, vendor-neutral serving layer lets organizations run open models, such as those in our open-weight roundup, with better throughput and lower cost.

AWS Emulator: Test Without the Cloud

On November 11, Amazon Web Services released AWS Emulator, a free local emulator for building and testing AWS applications on your computer or in continuous integration, without deploying or even needing an AWS account. Existing AWS SDK and CLI calls return in milliseconds, so developers and coding agents can find and fix errors offline.

Key details:

    • Services are generated from AWS's published API models.
    • Unsupported operations return a 501 Not Implemented response, so gaps are obvious.
    • Each instance gets its own port, keys and in-memory state.
    • There is no service fee, and the source is released under CC0 1.0.

For teams using coding agents, this is significant. An agent can test infrastructure code quickly and safely without touching real cloud resources or running up bills. It pairs naturally with GitHub's Agentic Workflows, which run agents in sandboxes.

Trajectory: Post-Training on Production Data

On November 2, Trajectory launched its self-serve platform for post-training AI models on production data, alongside a 45 million dollar Series A led by Sequoia. Teams connect their agents with a small SDK, plug in tasks with a way to score them and can then train, evaluate and deploy their own model, typically an open model, in one place without managing sandboxes or GPU infrastructure.

Two research previews launched with it:

    • Continual Grading, which learns graders from a few examples of expert judgment.
    • Group Size 1, which trains on real production interactions so a model keeps improving after it ships.

Harvey, the legal AI company, Rogo, Mercor and NVIDIA are named as early users. Harvey's involvement is a signal for the legal industry: specialized models trained on expert judgment are becoming practical. Permissively licensed bases such as Inkling 2 Small make this approach more accessible.

OpenAI Models on Google Cloud

Google Cloud added OpenAI's GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna to Vertex AI Model Garden in an invite-only private preview, usable via Google Cloud APIs and Codex workflows and payable with existing Google Cloud commitments, with general availability planned. Combined with Claude on AWS, Google Cloud and Azure, the major models are now available across all major clouds. For enterprises, that means choosing models based on capability rather than cloud contracts. Read more about GPT-6 Astra.

Other Notable Infrastructure Releases

    • Mistral's Responses API, which separates answers, reasoning and tool calls for easier agent loops, covered in our Mistral article.
    • Breken, a service that rebuilds your app in isolated worlds with synthetic users, aged data and failing dependencies, then reports reproducible bugs with traces and draft fixes.
    • llama.cpp 1.0, the stable release of the popular local inference library, covered in our local AI guide.

What This Means for Professional Firms

Lower Costs for AI Tools

Better serving infrastructure reduces the cost of running models, which flows through to the AI tools firms buy.

Specialized Models for Your Industry

Post-training platforms make it realistic for legal, accounting and healthcare software vendors to build models tuned to expert judgment. When evaluating vendors, ask whether their models are trained or graded on domain expertise.

Safer Automation

Local emulators and sandboxes let developers test automation without risking production systems. Firms that commission custom AI tools should ask developers how they test agent behavior before deployment.

Building Internal Tools

Firms with in-house developers can combine these tools with agent frameworks to build reliable automations. Our AI agent workflows, n8n templates and AI automation services show what is possible, and developers can sharpen skills with Claude Code for Python Developers.

Questions to Ask Your Technology Partners

    • Which models power the tool, and can we switch models later?
    • How is the model served, and what happens during traffic spikes?
    • How do you test agent behavior before releases?
    • Is the model fine-tuned or graded on expert work in our field?
    • Where is our data processed, and is it used for training?

Frequently Asked Questions

Is AWS Emulator free?

Yes. There is no service fee, and the source is released under CC0 1.0.

What is llm-d?

A CNCF-governed open-source core for distributed LLM inference, founded by Red Hat, IBM, Google Cloud, CoreWeave and NVIDIA.

Who uses Trajectory?

Early users named at launch include Harvey, Rogo, Mercor and NVIDIA.

The Bottom Line

The infrastructure behind AI agents is maturing quickly, making AI cheaper to run, safer to test and easier to specialize. Even if you never touch these tools directly, they shape the quality and cost of every AI product you use, so it pays to ask vendors about them. Explore our information technology AI resources for more.

Related AI News

Topics & Keywords
llm-d 1.0 AWS Emulator Trajectory post-training AI infrastructure LLM inference developer tools CNCF