AI coding agents can tend to isolate research, running experiments and generating ideas that are then forgotten when context windows reset. This can waste tokens, as models then repeat the same mistakes and hit the same dead ends.

But new research argues that it’s not the model itself, but the overarching ‘tree,’ that needs tweaking. To that end, data scientists from the Gaoling School of Artificial Intelligence, Renmin University of China, and Microsoft Research have introduced Arbor, a “persistent hypothesis tree” that helps agents remember and refine learnings over long research sessions.

A long-lived coordinator manages research strategy across the tree, while short-lived executors spin up isolated worktrees to test different hypotheses. As results come back, the tree updates, narrowing and refining throughout experimentation.

In practical tests, this technique delivered more than two-fold performance gains over standard AI coding agents across real-world engineering tasks, for the same budget.

This is because, said Mahmoud Ramin, a research director at Info-Tech Research Group, “Arbor accumulates information over time and allows agents to build upon prior discoveries just as humans do, through learning, adaptation, and eventually building upon what they have learned in the past.”

How Arbor grows

Arbor’s builders argued that longer execution on its own does not guarantee research progress. The challenge is maintaining a state that turns many individual attempts into “cumulative hypothesis refinement.”

Further, progress should not depend on human overseers regularly stepping in to dictate logical next steps or interpret the meaning of previous trials, they noted. To be truly autonomous, agentic research frameworks must maintain connections between experiments, data, results, and failures over time.

Arbor is built to fulfill three system requirements. First, it must be able to branch as sub-trees test out competing hypotheses that are all potentially plausible. At the same time, unrestricted branching can degenerate the whole framework, so that must be controlled to remain organized. The researchers call this “branching with coherence.”

Second, the infrastructure must separate local execution from overarching strategy. Testing out single hypotheses requires short-horizon tasks like editing, debugging, and evaluation. But these should not “obscure” the larger tree making decisions based on evidence gathered across the whole run.

Finally, the systems must be able to distinguish exploratory improvement from verified improvement. This prevents AI from overfitting during trial-and-error instead of iteratively learning from underlying patterns.

Persistence is at the core; the tree links hypotheses and ideas, the code or configuration artifacts used to test them, experimental evidence (results, metrics), and distilled insights (such as “this data filter helped, but this learning rate scheduler didn’t”).

Once a project kicks off, shorter-execution work trees run code, log their work, and collect metrics. The long-lived coordinator above them serves as the de-facto head of research, keeping an eye on the process, updating nodes, selecting “promising leaves,” pruning or merging branches, propagating reusable lessons, and deciding which hypotheses to pursue next.

“The tree therefore acts as the operational research state of the system,” Arbor’s builders wrote. “It is simultaneously the search frontier, the memory of past attempts, and the audit trail for verified artifact improvement.”

Outperforming Codex and Claude on new data

To test how well this process works, the researchers evaluated Arbor in an autonomous optimization (AO) setting: the agent was given an initial research artifact (a data pipeline, harness, or training script) and was tasked with improving its “held-out performance” through iterative experimentation, without human steering. Held-out performance is a machine learning (ML) metric that evaluates how well models are able to generalize on data they haven’t seen before.

The tree-based architecture was tested on several real research tasks across model training (its ability to improve training recipes and hyperparameters), harness engineering (how well it can upgrade evaluation or training harnesses), and data synthesis (its capacity to generate better data for training or evals).

Ultimately, Arbor outperformed the average held-out gains of Codex and Claude Code by 2.5x, for the same resource budget.

The takeaway, said the researchers: Keeping a structured, evolving hypothesis tree yields greater performance improvements than running the same models as ‘memoryless’ coding agents.

Arbor’s most innovative feature is its ability to maintain the agent’s memory and retain relevant data from prior attempts and hypotheses, Info-Tech’s Ramin pointed out, and, he said, “the next step for autonomous agents may be accumulating evidence over time.”

However, this does raise concerns about the auditability of robust research environments at a large scale, he noted. “As autonomous agents become more capable of performing work without human operators overseeing them, enterprises will need transparency into how and/or why an agent took a specific action or reached a certain conclusion.”

Enterprises implementing agentic AI face a challenge: Which tools should they allow their agents to use, where can they be found, and how can they be used safely? A new protocol, Agentic Resource Discovery, or ARD, aims to let agents answer those questions for themselves. Behind it are Google, Microsoft, Cisco, Nvidia, Salesforce and others.

ARD aims to standardize the way that tools and services are shared across systems within a corporate domain. For example, when investigating a production problem, an agent may want to query engineering documentation and open support tickets, deployment history and observability systems, all of which could be managed by different registries and across different silos. There is no common layer that pulls them together. ARD has been designed to be that layer.

It operates across two levels. Catalogs and Registries. In the first, an organization publishes a catalog setting out its available capabilities. The Registries layer act as a form of search engine, crawling those published catalogs.

The ARD specification is available now. Organizations are invited to publish their own catalogs using the quickstart guide. After this, they are able to join the community and participate in the evolution of ARD.

This article first appeared on InfoWorld.

Enterprises implementing agentic AI face a challenge: Which tools should they allow their agents to use, where can they be found, and how can they be used safely? A new protocol, Agentic Resource Discovery, or ARD, aims to let agents answer those questions for themselves. Behind it are Google, Microsoft, Cisco, Nvidia, Salesforce and others.

ARD aims to standardize the way that tools and services are shared across systems within a corporate domain. For example, when investigating a production problem, an agent may want to query engineering documentation and open support tickets, deployment history and observability systems, all of which could be managed by different registries and across different silos. There is no common layer that pulls them together. ARD has been designed to be that layer.

It operates across two levels. Catalogs and Registries. In the first, an organization publishes a catalog setting out its available capabilities. The Registries layer act as a form of search engine, crawling those published catalogs.

The ARD specification is available now. Organizations are invited to publish their own catalogs using the quickstart guide. After this, they are able to join the community and participate in the evolution of ARD.

With the market capitalization of AI companies soaring, US Senator Bernie Sanders is looking to give the American people a piece of the action.

The veteran senator for Vermont has introduced the American AI Sovereign Wealth Fund Bill, aiming to give the public a 50 percent ownership in the largest AI companies in the US. It’s a timely move with both OpenAI and Anthropic preparing for the imminent IPOs.

Sanders is not the only one pondering such a move. As the bill notes, OpenAI has proposed the creation of a “Public Wealth Fund” giving citizens a stake in “AI-driven economic growth,” while Anthropic has proposed a sovereign wealth fund to “shape the sector’s behavior.” President Trump’s advisors are also contemplating the possibility of the government grabbing a stake in major AI corporations, according to media reports.

Sovereign wealth funds are not a new idea: Many administrations across the world have implemented them, notably Norway which has about $2 trillion in its wealth fund.

Sanders’ proposal would give the public a direct ownership stake in the largest AI companies, but, he said, it wasn’t just about the wealth. “The foundation of AI is based on the collective knowledge of humanity and the creative work of tens of millions of people. The American people must have the ability to slow it down and make sure that AI benefits humanity, not just the richest people on the planet. That’s precisely what this legislation does,” he said.

An IT executive changing jobs usually attracts little attention outside a narrow group of people, but Noam Shazeer’s move from Google to OpenAI is as momentous as any high-value soccer transfer.

He announced the news in a post on X: “I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there.”

Shazeer initially achieved fame as one of the eight co-authors of the influential AI paper Attention Is All You Need, published when he was working at Google Brain. He is also one of the creators of the transformer technology that lies at the heart of modern AI models.

He left Google when the company failed to back his chatbot Meena and was tempted back when Google subsequently bought the company he founded, Character.AI, for $2.7 billion. That company achieved notoriety when it was sued by a grieving mother, who alleged that a Character.AI chatbot had contributed to her son’s death by suicide. The company subsequently settling out of court.

Shazeer has since been working as the co-lead on Google’s Gemini project. It’s not clear what role he will play at OpenAI, but hiring someone with his background shortly before the company’s IPO could be an attractive move for investors.

This article first appeared on Computerworld.

An IT executive changing jobs usually attracts little attention outside a narrow group of people, but Noam Shazeer’s move from Google to OpenAI is as momentous as any high-value soccer transfer.

He announced the news in a post on X: “I’m excited to share that I’ll be joining OpenAI and look forward to working with the exceptional team there.”

Shazeer initially achieved fame as one of the eight co-authors of the influential AI paper Attention Is All You Need, published when he was working at Google Brain. He is also one of the creators of the transformer technology that lies at the heart of modern AI models.

He left Google when the company failed to back his chatbot Meena and was tempted back when Google subsequently bought the company he founded, Character.AI, for $2.7 billion. That company achieved notoriety when it was sued by a grieving mother, who alleged that a Character.AI chatbot had contributed to her son’s death by suicide. The company subsequently settling out of court.

Shazeer has since been working as the co-lead on Google’s Gemini project. It’s not clear what role he will play at OpenAI, but hiring someone with his background shortly before the company’s IPO could be an attractive move for investors.

Google, Microsoft, OpenAI, and others want to help enterprises demonstrate that their AI applications are behaving themselves through the creation of a new foundation.

The Appia Foundation will, it explained rather impenetrably, “establish modular specifications that provide a connecting layer to bridge foundational global standards with practical, trusted assessments across the global AI value chain.”

Those specifications will help AI users ascertain whether the systems they are using meet all the obligations that apply to them in the form of standards and regulations, it said. It’s a challenging task with so much regional variation in requirements, and where the EU, for example, is more tightly controlled than the US.

The Foundation has established a set of criteria to demonstrate conformity with what is expected. There are two layers: the Requirements and Guidance layers will help users determine what is actually required, while the Assessment Enablement layer will look at how those requirements are evaluated.

Appia stressed that what it is offering are not standards — which are set by recognized international bodies such as ISO/IEC — but a means of assessing what those standards mean and how they can be used by organizations. However, the Foundation said that some of the criteria that it is introducing may become standards themselves after a period of time.

The Appia Foundation is hosted by the Linux Foundation’s Joint Development Foundation, and its other members include Arm, Ericsson, Mastercard, Mitsubishi Electric, Omron, Schneider Electric, and Siemens. It is also looking to bring academics and government into the fold, so that it can establish an advisory board.

This article first appeared on CIO.