Humans tend to be “a little bit precious about humans,” according to Eric Brandwine, distinguished engineer and VP at Amazon Security. 

We like to think we are all very good at our jobs, and we have high opinions of ourselves, he explained during a phone interview with The Register. “But when you actually get down to it, humans are not terribly consistent,” Brandwine said. 

Humans, like AI agents and systems, are non-deterministic. Neither can be guaranteed to produce the same output given the same input twice. Both will make mistakes and even make stuff up. However, we’ve got millennia of experience dealing with humans and less than a decade with more modern LLMs and the AI systems built on top of them. 

“We know how humans fail,” Brandwine said. “We’re comfortable with it. So human-in-the-loop isn’t necessarily the gold standard.”

For years, vendors have told companies that the solution for dealing with any automated system was to put a human in the loop. That battle cry became much louder with the advent of modern AI systems and reached a fever pitch when enterprises started deploying agents into their IT environments.

More recently, however, big tech is changing the way it talks about agentic governance and rethinking the whole human-in-the-loop concept.

Normalization of deviance

In 2017, Brandwine gave a talk on the normalization of deviance at AWS’ annual re:Invent conference. 

It’s a gradual process that happens when people in an organization take shortcuts, or don’t follow the established procedures or standards, and sometimes it occurs over years. As long as nothing catastrophic happens, this deviant behavior becomes the norm.

“It’s a thing all humans fall prey to, and one of the most heartbreaking stories I read in this area was about emergency departments and emergency rooms,” Brandwine said during a phone interview with The Register. “You’ve got all these machines, and they’re all beeping. Your first day on the job, you jump every single time one of the alarms beeps – but the patient is fine. It’s a spurious alarm. You go back to your station, you sit down, and over time, after enough of these false alarms, enough of these repeated beeps with no actual consequence, your discipline slips, and you stop responding. And eventually some tragic outcome occurs.”

This, he admits, is a very high-stakes example. And yet it’s a documented occurrence among healthcare workers, firefighters, and even Army pilots.

“Literally, someone’s life is on the line, and people still struggle to maintain discipline,” Brandwine said. “That’s the human condition.”

Here’s how this all applies to agentic AI governance and security. Humans build LLMs and AI systems, and having a “human-in-the-loop” ensures that a person reviews the AI’s output and approves (or not) any actions before the AI performs them.

“If you put a human inside of this tight loop, and ask them to make approval decisions for agentic tools repeatedly, time after time, they’ll do a good job,” Brandwine said. “And then they’ll do an okay job. And pretty quickly they’ll be doing a poor job.”

This is why at Amazon, “we’re not huge fans of human-in-the-loop,” he added. “It’s something that you should use judiciously, where you absolutely need it. But it’s not something that you can do at high velocity. You will not get the results that you want to get.”

Big tech pulls the human-in-the-loop

Amazon isn’t the first or only tech giant to start talking differently about the role humans should play in agentic governance. 

“It is very clear that we have moved from a human-led defense strategy, to a human-in-the-loop defense strategy, to an AI-led defense strategy that’s overseen by humans,” Google Cloud chief operating officer Francis deSouza told reporters during a press conference ahead of Google’s annual Cloud Next shindig in April. “Our model for the future is an agentic fleet that does a lot of the routine cyber security work at a machine pace and then is overseen by humans.”

Microsoft CEO Satya Nadella, in an X missive earlier this week, argued for “loop learning,” instead of having a human check an AI’s output at every step. 

“Companies need to turn their workflows, domain knowledge, and accumulated judgment into AI systems that improve with each use,” Nadella wrote. “Private evals should capture whether a model is actually improving against outcomes that matter to the business (not just external benchmarks!). Private reinforcement learning environments should let models grow stronger on real traces from inside the organization.”

Also this week, IBM execs called for human accountability – not humans in the loop – at all stages of AI development, deployment, and governance. 

Amazon’s alternative to human-in-the-loop is “accountability end to end,” according to Brandwine. This means human identity and ownership track through the entire workflow, even when humans aren’t directly approving every step.

“If I sit down at my keyboard and I type a command that takes a service down, I caused an outage,” Brandwine explained. “If I run a script that takes a service down, it’s still me that caused the outage. If my agent writes a script that they then run, and it causes an outage, that’s still my responsibility.”

(Secret) keys to the kingdom

This also highlights the importance of managing and securing agentic identities – the accounts, tokens, and credentials assigned to AI agents so they can access corporate apps and data. At Amazon, all of the agents have independent identities assigned to them, we’re told. 

“So, as we track agentic activity across our systems, it does not show up in the logs as: ‘Eric did this.’ It shows up as: ‘this agent did this on behalf of Eric,’” Brandwine said, adding that this isn’t to “make people afraid to use this technology.”

“It’s to make people pause and think: is this the right way to use this technology? Is this how I should be deploying this?” We still have the humans involved, we still have the humans making decisions, but we’re trying to play to the strengths of the humans rather than placing them in this unfair, repeated decision making, human-in-the-loop position.”

Brandwine told us that Amazon has run into a couple of hurdles when it comes to deploying agents across its businesses, and one of the biggest is what he calls “goal-seeking behavior.” This is when a person asks an agent to do a specific task – for example, upgrade a database – and the agent becomes laser-focused on just one action to achieve this goal, ie, deleting the database.  

This is separate from prompt injection because there’s no malicious input. “It’s just the agent getting stuck on the wrong action,” Brandwine said. Simply telling the agent, “you don’t have permission to do this,” is likely going to cause the agent to look for a different path to do the same thing (delete the database). 

Telling the agent why it doesn’t have permission to do something tends to produce a better outcome, according to Brandwine. This means telling the agent it’s not allowed to do that, and the reason why is because it would cause a production impact. And also include “don’t cause a production impact” as part of the prompt.

“Giving it that extra feedback has gotten us dramatically better results,” Brandwine said. 

Of course, this is not a fail-proof method. “You still need to be careful with agents,” Brandwine told us. “We have millennia of experience with humans. Agentic AI is a very, very new field, we don’t have an intuition for this, and one of the fundamental differences between agents and humans is that humans fear consequences,” such as losing a job or even going to jail. Agents don’t have these fears.

This is where setting permissions on what the agent can and can’t do or access comes in. Much like everything else with AI, it’s nuanced, and it depends on the employee’s role in the company, and the company’s tolerance for risk.

“The person that wants to run the agent wants to give the agent many permissions because that makes the agent more powerful,” Brandwine said. “It could do more things for them, it can recoup more of their time, it can deliver more.”

The security lead, on the other hand, wants to limit an agent’s permissions, and this causes yet more tension between the security and development teams. 

There is no one right solution or policy answer to solve this, according to Brandwine. Instead, it involves dynamic policies that set permissions based on the agent’s specific task.

There are some overarching, static guardrails – such as an agent must never perform destructive actions or delete entire servers – and then there are policies underneath that establish the maximum set of privileges that the agent can have.

“Then we’ll have a further scoped-down policy for this action, and there’s various techniques for automatically generating policies based on prompt and the end-user’s intent,” Brandwine said. 

Even for Amazon, it’s not always easy. “It’s all driven by risk,” he said. “This is a space that’s changing quickly, and so we’re trying to balance the risk of using untried, untested software against the risk of falling behind and not being able to deliver for our customers. As with all such things, it’s complicated.” ®

Earlier this week, at TechCrunch’s newest StrictlyVC event in El Segundo, Shinkei Systems founder Saif Khawaja and Founders Fund partner Delian Asparouhov sat down for a conversation that kept circling back to a question that doesn’t usually come up at a venture event: How do you know if a fish is stressed out?

It’s a fair question for Khawaja to field, since his company, Shinkei, has built its entire business around the answer. Shinkei makes a refrigerator-sized robot called Poseidon that fishermen install on their boats. The machine scans each fish with computer vision, identifies the species, and locates the brain. Within seconds of the fish coming out of the water, it pierces the brain and severs the gills, so the fish dies before it can thrash or suffocate.

That matters because a slow death floods the meat with stress hormones and lactic acid, which dull flavor and shorten shelf life. The whole thing is an automated, industrial-scale version of ike jime, a centuries-old Japanese technique traditionally performed dockside by trained fishermen at the moment of catch. By killing the fish instantly and draining its blood, ike jime delays decomposition long enough for the flesh to be safely aged for days, sometimes longer, before it’s served. That aging period is what gives top-tier sashimi its concentrated, umami-heavy flavor, as enzymes slowly break down the muscle.

Khawaja’s origin story is somewhat unusual for a hardware pitch. He grew up taking fishing trips with his family in the Middle East, and the idea Shinkei didn’t click until college, when he read an essay by an animal rights philosopher titled “If Fish Could Scream.” Its premise was that fish lack vocal cords, so the suffering most of them experience on the way to your plate is essentially invisible. Conventional commercial fishing typically lets fish suffocate on deck, a process that can take anywhere from a few minutes to roughly an hour. During that time, fish release stress compounds that shorten shelf life and dull flavor, the same basic mechanism that makes a stressed cow produce tougher, less flavorful beef.

But Shinkei’s ambitions have expanded well past the killing machine. The company now describes itself as a vertically integrated fish harvester and processor, deploying robotics and AI across the chain from boat to plate. Shinkei gives Poseidon machines to fishermen for free, then pays those fishermen a premium price for the fish that come out of them, well above what the catch would fetch at a standard dock auction. In exchange, Shinkei takes full possession of the fish rather than letting fishermen sell it on the open market. The catch then ships to a 16,000-square-foot plant Shinkei bought in Tacoma, Washington, where it’s broken down and sold under the company’s consumer brand, Seremoni, marketed as “ceremony grade” fish.

The most visible proof point so far is on the menu at Erewhon, the Los Angeles grocery chain beloved by influencers. Erewhon sells Shinkei’s fish as Seremoni Grade Miso Black Cod, hot off the prepared-foods bar, and the marketing around it leans hard on the “sustainably caught, humanely harvested” framing. The arrangement is still a pilot, running for now out of Erewhon’s Manhattan Beach location, with wider rollout to other stores contingent on how well it sells. Khawaja says the company already supplies fish to restaurants holding a combined 50 Michelin stars, and claims something that has reportedly never happened before: Japan importing American-caught fish into its own fish markets, which have historically treated American seafood as distinctly inferior to the domestic product.

Whether buyers will pay a premium for “humanely killed” fish, the way many now do for humanely raised beef and poultry, is still an open question, and even Khawaja treats it as secondary to the pitch when asked about this. He told the El Segundo crowd the real selling point isn’t the animal-welfare story so much as the practical one. A catch that might normally have a 5-to-7-day shelf life can stretch to 12 or 14 days, he said, and the company has cooked fish three weeks after coming out of the water with no issue. Shinkei’s newest product, an in-plant sensor system, tries to quantify that by scanning fish and projecting an individual shelf life for each one. That matters in an industry where, by Khawaja’s estimate, roughly 18% of product is lost to spoilage just between dock and store, before retail loss is even counted.

That spoilage problem is tangled up with a detail of the American seafood supply chain that surprises most people who haven’t worked in it. A meaningful share of fish caught in U.S. waters by U.S. boats gets frozen and shipped overseas, often to China, for the labor-intensive work of heading, gutting, scaling and filleting, then shipped back to be sold here. Industry estimates of how much American seafood is imported run as high as 90%, though roughly half of that, by some estimates, actually originated in domestic waters before making the round trip abroad. Reporting has tied parts of China’s seafood processing sector to forced labor, including Uyghur workers in Shandong province and North Korean labor in Liaoning, making the system a target of U.S. trade and labor scrutiny in recent years. There’s been a push within the industry to “re-shore” some of that processing, spurred partly by tariffs and pandemic-era disruptions that made the China round trip less attractive.

The bet that Shinkei — and Founders Fund — are making is that re-shoring the entire chain, catch, kill, process, and distribute, all under one roof in Tacoma, can be done profitably enough to outcompete it.

For Founders Fund, the wager fits a pattern the firm has leaned into for years, which is backing founders who are often outside of fashionable categories. Asparouhov, who speaks a mile a minute and without reserve, put it plainly: there’s essentially nobody else on Earth who wants to spend their life on robots that kill fish, and the smell of the office makes that clear enough. (It was a very funny line, though it undersells the field a little. In addition to Shinkei, a Japanese firm called Nichimo sells a device that stuns fish to assist humans performing ike jime by hand, and several Norwegian startups are building robotic systems for more humane fish slaughter and processing. Shinkei’s edge, for now, is being the only one running the fully automated version of the technique at scale on U.S. boats.)

In fact, Asparouhov said the firm intentionally keeps its exposure to crowded categories like generic AI applications relatively low. By his rough math, AI and defense together account for something like 15% to 20% of the fund’s deployed capital, well below what he estimated is typical elsewhere in venture. Shinkei sits alongside Halter, a New Zealand-founded company making solar-powered, GPS-equipped cattle collars that let ranchers herd cattle remotely, and Ohalo Genetics, the crop-genetics company started by “All-In” podcast co-host David Friedberg, as evidence that the firm’s appetite for food and agriculture isn’t a one-off.

Of course, the fund’s headline-grabbing recent win has nothing to do with fish. Its early and aggressive bets on Elon Musk’s SpaceX — a relationship that traces back to Peter Thiel and Musk’s shared history at PayPal — are reported to have generated tens of billions of dollars for the firm, by some accounts the largest venture outcome ever recorded. Asparouhov argued that win has accelerated a broader shift in venture toward hardware and physical-world businesses, noting that most of the largest companies on the Nasdaq today involve complex electromechanical systems rather than pure software. He predicted more of SpaceX’s alumni, flush with liquidity and shaped by working alongside Musk, will go on to start their own ambitious physical-world companies.

Whether Shinkei becomes one of the firm’s next big wins will take time to know. The company is a robotics manufacturer and a seafood processor and a consumer brand, all running at once, and each layer has its own daunting challenges. Fishermen are used to working a certain way. Distributors are built around decades-old habits. Chefs and grocery buyers still have to be convinced that a story about humane fish slaughter is worth paying more for. The hardware has to survive saltwater, fish guts, and life on a commercial boat, and the product it’s selling spoils, so there’s little room for the kind of stumble a software company can usually shrug off.

Still, talking with the two together in El Segundo was enough to make me understand why Founders Fund finds the bet compelling. The firm thinks it has found a founder building something novel in a surprisingly dysfunctional industry — the kind of company almost nobody else in the United States even wants to build.

Save $76: Amazon’s Prime Day kicks off on June 23, but the deals are already in bloom. The retailer just slashed the price of Nothing’s wireless earbuds by $76, dropping them to an all-time low. This is a limited-time deal available only to Prime members, so we suggest acting fast.

Finding a premium pair of earbuds with long battery life, active noise cancellation and fast charging can normally be quite steep. However, brands like Nothing offer an incredible, budget-friendly alternative without sacrificing great sound.

Nothing earbuds include eartips in three sizes for a perfect fit and the most out of their features. They offer up to 45 decibels of noise cancellation and have a battery that lasts up to 40 hours. The 11mm drivers provide clear bass and treble for an immersive listening experience, no matter the music genre or audiobook. If you’re an avid ChatGPT user, you’ll enjoy Nothing’s integration with the popular AI service. Simply set up integration using the Nothing app, and you can use voice controls to experiment with it.

Nothing also improved the earbuds’ connectivity, making it easier to switch between devices whenever you need to pick up on a quick phone call, join a video meeting and more. Thanks to HiRes Wireless Audio, your voice will be heard loud and clear during phone calls, so you won’t have to worry about any distractions interrupting family or work discussions.

We’ve compiled a list of the best earbud and headphone deals so you can find exactly the right discount for you.

Alternatives you might like

Nothing earbuds include noise cancellation, crisp sound and AI features at a new record-low price. However, if these aren’t your style, we’re huge fans of Apple’s AirPods 4, which made our list of best wireless earbuds of 2026. You can snag the ANC model for $149, down from $179. Or grab the pair without ANC for less than $100 and save 23%.

The AI Act, which entered into force in August 2024, attempts to tackle some of the risks emerging from the technology while letting the bloc benefit from its economic potential.

A bill to enforce the EU’s AI Act in Ireland has been approved. Once enacted, the Regulation of Artificial Intelligence Bill 2026 will establish ‘Oifig IS na hÉireann’ – or the AI Office of Ireland – as an independent statutory body which will act as Ireland’s central coordinating authority to implement the landmark EU legislation on AI.

The AI Act, which entered into force in August 2024, attempts to tackle some of the risks emerging from this technology while letting the bloc benefit from its economic potential.

The law applies in a risk-based and phased manner across all EU member states, imposing obligations on providers, deployers and importers of AI systems and models. Last month, the EU published draft guidelines for what it considers ‘high-risk’ AI systems.

Meanwhile, the new bill approved by the Government is only a technical regulation needed for the supervision and enforcement of the EU AI Act and does not add to Ireland’s existing obligations. It comes as Ireland readies to assume presidency over the Council of the European Union from 1 July to 31 December this year.

The Irish bill provides market surveillance authorities (MSAs) in the country with an enforcement toolkit for the AI Act, enabling them to issue compliance notes and fines, or even prosecute entities. As one of the MSAs, the Irish Competition and Consumer Protection Commission is also introducing a new general administrative sanctions procedure.

“AI is a transformative technology which offers extraordinary potential for our economy and citizens, but requires appropriate oversight and accountability to ensure people are protected,” said Minister for Enterprise, Tourism and Employment Peter Burke, TD.

“This bill delivers this approach. It fulfils Ireland’s EU obligations, giving effect to pioneering AI regulation in domestic law, while ensuring we have the national infrastructure to enforce it effectively.

“The bill establishes the AI Office of Ireland as a strong, independent institution at the centre of our AI regulatory system and empowers our competent authorities with the investigative and sanctions tools they need.”

Minister of State for Trade Promotion, Artificial Intelligence and Digital Transformation Niamh Smyth, TD said: “This bill is about more than regulation. It is about building the institutional foundations for a future in which AI works for people, ethically, transparently and accountably.

“The establishment of the AI Office of Ireland will give us a world-class focal point for AI regulation, innovation and expertise. Ireland is a key player in the global AI ecosystem, home to many of the world’s leading foundational AI model providers.”

While broadly considered to be a first of its kind, the EU AI Act faces a number of emerging challenges in its implementation, one of them being the launch (and subsequent blockade) of Anthropic’s Mythos and Fable models in Europe, which prompted experts to question how the Act might control risks emerging from foreign-created and deployed AI technology.

The bloc, on the other hand, is also attempting to remedy some of the AI law’s supposed over-regulation by way of a simplified and consolidated set of rules.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Imagine your engineering team just deployed an AI agent to search through internal company documents and answer employee questions. It works perfectly in development, but in production, it consistently hallucinates or misses key constraints. Fixing this is rarely a simple patch. It requires a tedious, trial-and-error process of tweaking chunking strategies, retrieval methods, and system prompts simultaneously. Because these adjustments are entangled, it becomes nearly impossible to attribute which specific tweak actually solved the problem. 

To address this challenge, researchers at Renmin University of China and Microsoft Research introduced Arbor, a framework that upgrades AI-driven research and optimization from a sequence of trial-and-error guesses into a cumulative learning process. Arbor organizes hypotheses, experiments, and insights into a tree that helps the system learn from prior failures to make smarter, verified improvements over time.

In practical tests, Arbor delivered more than 2.5 times the verifiable performance gains of standard AI coding agents across real-world engineering tasks while operating under the same resource budget. 

For enterprise AI, this technique directly translates to automating the continuous improvement of complex, real-world engineering systems.

Understanding the bottleneck in autonomous optimization

As large language models and AI systems become more capable, they are expected to carry out more complex operations such as autonomous optimization (AO) of software systems such as agent harnesses or model training algorithms. 

AO captures the fundamental loop of autonomous research. An AI agent starts with an initial mutable artifact, such as a machine learning codebase or data pipeline, and a specific objective. The agent’s goal is to iteratively improve this artifact through experimental feedback without step-by-step human supervision.

The main challenge of AO is often misunderstood. Many engineering teams find that simply giving a coding agent more time or compute to optimize a codebase doesn’t lead to better results. “Automation can keep an AI working for a very long time — but a loop is not the same as progress,” Jiajie Jin, co-author of the paper, told VentureBeat. “If the goal is vague, or the metric is easy to hack, long-running automation often just produces ‘improvements’ faster that nobody actually wants.”

Jin explains that complex tasks take many attempts to get right, and standard agent architectures are missing the critical data structure to maintain state. “How do you make sure the insight and experience from each attempt actually accumulate, instead of getting lost in a scrollback buffer?” he said. Without this structure, agents simply repeat the same mistakes.

Current agent systems can run experiments for many hours against well-specified goals: editing code, invoking tools, running tests autonomously. But they treat each attempt in isolation, missing the structural mechanisms that would let them accumulate and act on what they’ve learned.

They lack the capacity to simultaneously maintain and compare multiple competing research directions. Without this, they cannot interpret both successes and failures to reshape their future exploration, which is the core mechanism that makes human research cumulative.

General coding agents typically rely on conversation transcripts for their memory. Because AO tasks span hundreds of turns and easily exceed context window limits, these agents struggle to preserve and reuse factual evidence over long histories. As a result, they lose the overarching structure of the research process and are prone to stalling on early failures or chasing noisy evaluation swings. The system needs a structured, durable memory that records what directions have been tried, what factual evidence was produced, and how each result changes the space of future hypotheses.

Existing frameworks are also prone to reward hacking and overfitting to development metrics. This makes them create the illusion of progress without producing improvements that transfer to real-world performance.

Finally, general-purpose coding agents typically chain their tool calls on a single shared working tree. This architectural limitation prevents them from testing parallel hypotheses in isolated environments without corrupting the main codebase or obscuring which hypothesis caused a specific outcome.

The Arbor framework

Arbor solves the challenges of AO with a framework that automates the long-horizon loop of exploration, experimentation, and abstraction that characterizes human research. Arbor separates the strategic direction of research from the ground-level coding tasks with two key components:

The coordinator: A long-lived AI agent that acts like a principal investigator. It never directly edits the target codebase. Instead, it owns the general state of the optimization research, observes accumulated evidence, comes up with new hypotheses and directions to explore, and decides what to do with the results of experiments.

Executors: Short-lived, highly focused AI agents. When the coordinator wants to test an idea, it spins up an executor and places it in an isolated environment, essentially a fresh git worktree. Each executor is handed one hypothesis. It implements the assigned idea, runs evaluations, debugs errors, and reports back to the coordinator with the results and created artifacts.

These two components collaborate through a mechanism that the researchers call “Hypothesis Tree Refinement” (HTR). HTR represents the entire research process as a persistent, branching tree where every node binds together four things: a hypothesis, the executable artifact, the factual evidence produced, and a distilled insight. This means the coordinator can explore multiple competing directions at the same time without losing its place.

The coordinator builds the tree by placing broad ideas near the root, while concrete refinements branch out as leaves. This allows Arbor to safely explore multiple competing hypotheses simultaneously. If an executor’s experiment fails, the tree records why it failed as a negative constraint, ensuring the system doesn’t endlessly repeat the same mistake.

To understand why Arbor’s isolation matters, consider a common enterprise scenario: optimizing a Retrieval-Augmented Generation (RAG) pipeline for an internal AI assistant. “When you ask a single agent like Claude Code or Codex to ‘improve accuracy,’ it will typically change a bunch of things in one pass — chunking, the prompt, the retrieval method,” Jin said. This entangles the changes, making it impossible to attribute which one actually helped. It also directly mutates the repository without isolation. 

Arbor solves this by treating each lever as a separate hypothesis. Chunking becomes one branch, retrieval another, and the prompt another — each implemented and evaluated in its own isolated git worktree. “So you get clean attribution: ‘constraint decomposition on the retrieval side gave +X; breadth-first search actually hurt,’” Jin said.

When an executor returns a report, the coordinator writes the evidence to the tree and backpropagates the insight upward to parent nodes. This means a local observation becomes a generalized constraint that shapes the coordinator’s future idea generation.

To prevent reward hacking or overfitting to the development data, HTR enforces a strict “merge gate.” Even if an executor reports a fantastic development score, the coordinator will spin up an isolated worktree to test the candidate against a held-out test evaluator. The artifact is only merged into the current best trunk if it demonstrably improves the test score, verifying that the progress is real.

Arbor generally falls under the concept of “loop engineering,” popularized by industry figures like OpenClaw creator Peter Steinberger and Claude Code lead Boris Cherny. The idea is to move beyond single prompts to design iterative cycles (observe, reason, act, verify) that drive autonomous agents. However, as Jin points out, “A loop can fill up with messy, untraceable attempts, and you end up with nothing to show and no way to reconstruct what changed.”

Arbor in action

The researchers evaluated Arbor on an autonomous optimization task suite built from real-world research settings and the MLE-Bench Lite machine learning engineering benchmark. The AO suite featured tasks from different areas of AI development, including model training, harness engineering, and data synthesis.

The researchers used different backbone models for the coordinator and executor agents, including Claude Opus 4.6, GPT-5.5, and Gemini-3-Flash. They tested Arbor against the strongest coding agents, Codex and Claude Code. Arbor and the baselines were given the same resources. For the MLE-Bench Lite tasks, Arbor was also compared against top-tier agentic research systems like AI-Scientist, ML-Master, and AIDE.

Arbor consistently outperformed the baselines. It achieved the best held-out test result on all tasks, attaining more than 2.5 times the average relative gain of Codex and Claude Code. On the BrowseComp task, which involves optimizing a search agent, Arbor improved the system’s held-out accuracy from a baseline of 45.33% to 67.67%. Meanwhile, Codex and Claude Code stalled at 50% and 53.33%, respectively. On MLE-Bench Lite, when equipped with GPT-5.5, Arbor achieved the strongest result among all benchmarked systems.

Arbor proved to be resilient against overfitting. For example, during the Terminal-Bench 2.0 task experiments, Claude Code achieved a high development score of 75 but its score dropped to 71 on the held-out data. Arbor had a lower development score of 72.22 but achieved the highest held-out score of 77.36, ensuring its results transfer to real-world applications.

Arbor also showed generalization in a cross-task transfer experiment. After Arbor finished optimizing the search harness for the BrowseComp task, researchers took the optimized codebase and tested it on two unrelated search-agent tasks, HLE and DeepSearchQA. Arbor’s optimized codebase significantly improved performance on those unseen tasks as well.

Deploying Arbor: Sweet spots and hidden costs

For engineering leads looking to drop Arbor into their existing tech stack, the framework is designed to sit on top of existing Git workflows rather than replacing them. “Its output is an ordinary git branch that your existing code review, CI, and human review can inspect directly,” Jin said. Only verified gains are merged into a per-run trunk, leaving the main repository untouched until a developer manually chooses to promote the code.

However, deploying Arbor comes with specific tradeoffs. Jin points out that the biggest catch is token cost, as maintaining a long-lived coordinator that continuously manages the tree and dispatches executors is the dominant expense. Running multiple isolated worktrees concurrently also requires genuine compute and disk resources to process real experiments.

So where is Arbor’s sweet spot? According to Jin, it excels at tasks with a clear, trustworthy metric, tolerance for a long time horizon, and a real search space with several plausible directions, such as pipeline optimization, data-synthesis quality, and model-training recipe tuning. 

Conversely, teams should explicitly avoid using Arbor for real-time latency tasks, obvious one-line fixes, or when the underlying evaluation metric is flawed. The quality ceiling of the entire run is strictly bounded by the quality of the evaluator. “If the metric isn’t trustworthy, Arbor will just optimize toward an untrustworthy result faster,” Jin said.

Jin sees the next evolution going beyond single scalar metrics. “A natural evolution is to have each node’s artifact carry a vector — accuracy, latency, cost — instead of a single score,” Jin said. “Going from a single scalar to a multi-objective Pareto search is a very natural extension of the framework.”

John Jumper, who shared a recent Nobel Prize in chemistry, announced Friday that he’s making the leap to Anthropic after “nearly 9 years” at Google DeepMind.

In a post on X, Jumper wrote that DeepMind CEO Demis Hassabis “took a real chance letting me lead the AlphaFold team just six months after finishing my PhD, and the entire GDM team taught me so much about how to do great science.”

Jumper (pictured above right, with Hassabis) added, “GDM is a special place, and I’ll still be excited to hear about what amazing things they discover next.”

Bloomberg reports that Jumper was a key member of Google’s team developing coding tools, which the company has struggled to sell to businesses. Character AI co-founder Noam Shazeer also announced this week that he’s leaving DeepMind — though in Shazeer’s case, he’s joining OpenAI. 

Jumper and Hassabis won the Nobel Prize in 2024 for their work on AlphaFold, an AI model that can predict the 3D structure of proteins based on their genetic sequences.

Following the May 2026 launch of its flagship Velsonic phono preamp, AVID HiFi is expanding its analog lineup with the Pulsus II and Pellar II phono stages. Both models replace long-serving originals: the Pulsus, introduced in 2010, and the Pellar, which followed in 2012. That is a respectable run in any product category, but especially one built around a format that some people were still declaring dead when those first models arrived.

The vinyl train keeps rolling, and there is no sign that anyone has found the brakes. In 2026, buyers have more turntables, phono cartridges, phono preamps, record-cleaning machines, tonearms, cables, mats, clamps, and other playback accessories to choose from than at any point in the past two decades. AVID’s updated Pulsus II and Pellar II arrive in a market that is no longer treating vinyl as a nostalgic side hobby, but as a serious and increasingly sophisticated part of the high-performance audio ecosystem.

On the inside, both models retain much of what made the originals successful, with targeted performance refinements wrapped in a cleaner, more streamlined exterior. AVID is not a company that releases new models every year simply because it has something new to promote. The original Pulsus and Pellar earned considerable recognition for their musicality, engineering integrity, and long-term reliability—qualities that kept both phono stages relevant for well over a decade.

That kind of longevity suggests that AVID got the fundamentals right from the beginning. The Pulsus II and Pellar II are therefore not reinventions for the sake of it, but the result of genuine advances in design, materials, and the company’s deeper understanding of how these circuits can be improved.

AVID Pulsus II and Pellar II: Refined, Not Reinvented

The Pulsus II and Pellar II build on the foundations of their predecessors with refinements in several critical areas. AVID points to more careful component selection and improved noise-reduction strategies, intended to deliver greater transparency, lower noise floors, stronger dynamic expression, and a more natural presentation of music. Both models also offer flexible adjustment options for Moving Magnet (MM) and Moving Coil (MC) cartridges.

As noted earlier, the Pulsus II and Pellar II arrive shortly after the launch of AVID’s flagship Velsonic phono preamp. Together, the three models give serious vinyl listeners a broader range of options based on their cartridge requirements, preferred feature set, and budget.

AVID’s expanded three-model phono preamp lineup reflects the company’s continued focus on meaningful engineering rather than annual cosmetic updates. The Velsonic, Pulsus II, and Pellar II mark a new phase for AVID Hi-Fi in the phono-stage category, with choices that extend from more attainable high-performance designs to a no-compromise flagship.

Plusus II and Pellar II Differences: Highlights

Detailed Comparison

The Bottom Line 

The AVID Pellar II and Pulsus II are not budget phono preamps with a nicer faceplate. Both are serious, UK-built MM/MC phono stages aimed at vinyl listeners who have moved beyond the built-in phono input on an integrated amplifier and want lower noise, better cartridge matching, and a more substantial long-term analog upgrade.

What makes AVID’s new pair interesting is the company’s refusal to treat either model as a disposable annual refresh. These are successors to designs that remained in service for more than a decade, now updated with revised component selection, lower-noise engineering, and a cleaner industrial design. The Pellar II is the more accessible entry point for listeners with one well-chosen turntable and cartridge, while the Pulsus II is the step-up option for systems where the phono stage is expected to extract more from a serious moving coil cartridge and a higher-resolution analog front end.

Competition is not exactly thin. The $2,495 MoFi UltraPhono Pro offers front-panel gain and loading adjustments, balanced and single-ended outputs, and a very different feature set. The $2,899 Cyrus 40 PPA adds four configurable inputs, balanced outputs, remote control, and upgrade potential for listeners running multiple turntables or tonearms. The tube-based Muarah MU-2 remains another compelling option around the Pellar II’s price, particularly for listeners who want easy cartridge loading changes and a warmer, more relaxed presentation.

The obvious question is whether the $4,695 Pulsus II delivers enough of a performance advantage over the $2,295 Pellar II to justify its substantial premium. That answer will depend heavily on cartridge quality, system resolution, and how deeply invested the buyer is in moving coil playback. For most vinyl listeners, the Pellar II is likely to be the more rational AVID entry point. The Pulsus II is for owners of more ambitious analog systems who are willing to pay considerably more to chase lower noise, greater refinement, and a higher ceiling.

And yes, there are far less expensive ways into external phono preamps. The iFi ZEN Air Phono 2 remains a sensible low-cost MM/MC option, but it belongs in a different conversation.

Price & Availability

The AVID HiFi Pulsus II and Pellar II phono preamps are designed and manufactured in the UK. Both will be available through authorized AVID dealers worldwide beginning July 2026. 

• The Pulsus II is priced at £2,995 / €3,995 / $4,695.
• The Pellar II is priced at  £1,450 / €1,995 / $2,295.

Related Reading: