Academic rigour, journalistic flairSearch analysis, research, academics…

Arts + CultureBusiness + EconomyEducationEnvironmentHealthPolitics + SocietyScience + TechWorldPodcastsInsights

Share article

Print article

The UK government has just approved the Jackdaw gasfield in the North Sea, clearing the way for production to begin as early as this winter.

Jackdaw, around 150 miles east of Aberdeen, could account for roughly 6.5% of UK North Sea gas production at its peak and will connect to existing infrastructure at the Shearwater platform.

But what will this actually mean for households and public finances? Will Jackdaw lower energy bills? Probably not by any noticeable amount.

---

Read more: Granting a new gasfield licence does not guarantee energy security

---

Producing more British gas may sound as though it should lower prices. But North Sea gas is not reserved for British consumers at a special domestic price. It is sold into a market connected to Europe and influenced by international gas prices. This was clear following Russia’s invasion of Ukraine. Britain continued producing gas domestically, yet consumers were still exposed to soaring international wholesale prices.

According to one analysis, even if Britain maximised North Sea production and returned all resulting government revenues to households, the average annual saving would be no more than £82. Jackdaw is one field, so its effect would be small.

More than 84,000 people start the day with our Daily Email.

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Will this new gas field licence improve energy security? Here the case is stronger.

Britain consumes more gas than it produces and increasingly depends on imports as older North Sea fields decline. Jackdaw will not make Britain energy independent, but additional domestic production can reduce imports.

There is an important distinction between security of supply and protection from high prices. Jackdaw can contribute to the first, but not the second while consumers remain exposed to international gas prices.

In terms of how much tax Jackdaw will generate, North Sea taxes are currently charged on profits, not total sales. Investment reliefs (tax breaks that allow companies to deduct certain investment costs) can reduce those taxable profits.

Companies can generally deduct day-to-day operating costs when calculating taxable profits. They can also claim tax relief for certain development costs, investments to reduce emissions and, eventually, the costs of dismantling the field.

For Jackdaw, development investment will be in the region of £2.1 billion according to Adura, the UK North Sea’s largest independent oil and gas producer. Yet using the developer’s production forecast which estimates of how much gas and other products Jackdaw will produce each year, and the UK government’s Office for Budget Responsibility’s March 2026 oil and gas price assumptions, we estimate Jackdaw would generate only around £1.5 billion to £1.6 billion in gross sales by the end of 2029. Around £1 billion would come from gas, with the remainder from condensate and natural gas liquids (these are the liquid products extracted alongside natural gas that are sold separately).

Based on this simple comparison, gross sales through 2029 would still be below the project’s implied development investment, even before operating costs are taken into account.

This does not mean Jackdaw will pay no tax. Public information does not reveal how much investment qualifies for tax relief, when allowances will be used or how Jackdaw fits into the operator’s wider North Sea tax position. Adura estimates that Jackdaw and the much larger Rosebank project together could generate around £1.4 billion in tax by mid-2029 but does not say how much of that would come from Jackdaw.

There is another complication. The energy profits levy (a temporary 38% tax on oil and gas profits) is due to end by March 2030. It will be replaced by the oil and gas revenue levy, a new tax that would apply when oil and gas prices rise above specified levels. Unlike the current windfall tax, this would charge 35% on revenues above specified price thresholds, regardless of a field’s costs. My research suggests the same revenue threshold can affect high- and low-cost fields very differently.

What Jackdaw really changes

Jackdaw will not give Britain cheap gas or make the country energy independent. Its significance is more limited: it can replace some imported gas and support economic activity, but Britain will remain exposed to international gas prices.

The financial picture is similarly complicated. A 78% headline tax rate does not mean 78% of Jackdaw’s revenues will reach the Treasury. Investment reliefs can reduce tax in a field’s early years, while jobs and supply-chain activity generate other tax revenues and eventual decommissioning costs can also qualify for tax relief.

These benefits must be weighed against the emissions and environmental costs of further fossil-fuel production, and against investing instead in renewables, storage and other ways of reducing dependence on gas.

By producing a little more of its own gas, Britain can marginally improve its security of energy supply, but this won’t lower prices or guarantee an immediate tax windfall.

• Energy security
• Fossil fuels
• Oil and gas
• North Sea
• Energy bills
• Oil and gas fields

Authors

1. Karl MatikonisAssistant Professor in Taxation, University College Dublin
2. Hafez AbdoAssociate Professor of Accounting, University of Nottingham

Disclosure statement

The authors do not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and have disclosed no relevant affiliations beyond their academic appointment.

Partners

University of Nottingham provides funding as a founding partner of The Conversation UK.

University College Dublin provides funding as a member of The Conversation UK.

The Conversation UK receives funding from these organisations

View the full list

DOI

https://doi.org/10.64628/AB.r5s4mgve4

We believe in the free flow of information

Republish our articles for free, online or in print, under Creative Commons licence.Republish this article

Help us help the climate

Climate action is stalling, and our information ecosystem is one of the reasons why. The Conversation cuts through the noise with climate reporting direct from the climate scientists. If you value our approach, please support our not-for-profit newsroom.

Donate now

Anna Turns

Senior Environment Editor

61,507 views Dec 12, 2024 Robert Sapolsky Father-Offspring Interviews

Episode 41 of Father-Offspring Interviews. This episode discusses the role of sleep in depression, whether Bipolar Disorder is a sleeping disorder, and the psychosomatic nature of chronic pain. Captions available for translation.

Conversational AI Is Cracking the Neuroimaging Bottleneck

FeaturedNeurologyNeuroscience

October 21, 2026

Summary:

A multi-institutional team led by Lehigh University has received a four-year, $2.8 million NIH grant to develop UniBrain, an all-in-one artificial intelligence platform designed to consolidate fragmented neuroimaging pipelines. By unifying preprocessing, segmentation, multi-modal integration, and disease prediction into a single deep learning architecture, paired with a conversational agentic interface, UniBrain aims to turn raw MRI scans into interpretable diagnostic biomarkers in minutes rather than months.

Key Facts:

• $2.8M NIH Initiative: Funded by the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the four-year project is spearheaded by Lehigh University in collaboration with Massachusetts General Hospital/Harvard Medical School, Emory University, and the University of Pittsburgh.
• Unified Pipeline Architecture: UniBrain eliminates the need to shuffle brain imaging data across disparate, specialized software packages by combining image registration, tissue segmentation, feature extraction, and network-level disease prediction into one cohesive system.
• Agentic AI Interface: The system incorporates a conversational, ChatGPT-like interface that allows clinicians and researchers without computational backgrounds to upload raw MRI files, ask diagnostic questions in plain language, and receive automated, interpretable biomarker reports.

Source: Lehigh University

Extracting clinically actionable biomarkers and neuroscientific insights from brain scans remains one of the most computationally tedious bottlenecks in medicine. Currently, translating raw neuroimaging data into diagnostic predictions requires researchers to shepherd massive datasets through a fragmented sequence of specialized, single-purpose software packages, one for motion correction, another for anatomical registration, a third for tissue segmentation, and separate tools for network modeling.

For extensive clinical cohorts, this manual pipeline can take weeks or even months to execute, creating a substantial barrier for clinicians and neuroscientists who lack advanced coding and computational infrastructure.

To solve this fragmentation, an interdisciplinary consortium led by Lifang He, PhD, associate professor of computer science and engineering at Lehigh University, has been awarded a four-year, $2.8 million grant from the National Institute of Biomedical Imaging and Bioengineering (NIBIB) to develop UniBrain, an artificial intelligence platform engineered to consolidate the entire neuroimaging workflow into a single, automated architecture.

Consolidating Fragmented Neuroimaging Workflows

The initiative brings together computer scientists, radiologists, bioengineers, and neuroscientists across four leading research centers: Lehigh University, Massachusetts General Hospital/Harvard Medical School (led by Xiang Li), Emory University (led by Carl Yang), and the University of Pittsburgh (led by Liang Zhan).

Traditional pipelines handle key computational stages, such as spatial normalization, skull stripping, volumetric segmentation, structural/functional connectome mapping, and pathology classification, as isolated, disconnected tasks. This modular approach compounds error propagation, requires extensive file format conversions, and demands manual quality assurance at every transition.

UniBrain re-engineers this pipeline from the ground up by deploying end-to-end deep learning models capable of simultaneous preprocessing and downstream neural network analysis. Crucially, the platform harmonizes complementary data across multiple imaging modalities, including structural MRI, functional MRI (fMRI), and diffusion tensor imaging (DTI), within a unified latent space.

Preliminary tests demonstrate that UniBrain’s integrated framework matches or exceeds the precision of conventional multi-tool pipelines while reducing computational overhead from weeks to minutes.

“This is a very ambitious project,” said Dr. He. “We’re developing and deploying new AI tools to do the analysis while rethinking the entire neuroimaging analysis process from start to finish within a unified platform. We need experts in many fields working together to solve problems of this scale.”

An Agentic, Conversational Interface for Medicine

Beyond building the computational backend, the consortium is developing an agentic AI framework designed to democratize advanced neuroimaging for frontline medical professionals.

Operating through an intuitive, conversational interface modeled after large language model assistants, the agentic system allows physicians and neuroscientists to upload raw DICOM or NIfTI imaging files and interact with the data using natural language prompts. Clinicians can query specific anatomical anomalies, request automated regional volume comparisons, or track functional connectivity changes over time.

In response, UniBrain autonomously selects the optimal preprocessing parameters, runs the underlying deep learning classifiers, and generates transparent, interpretable clinical reports complete with visual saliency maps highlighting subtle disease biomarkers that human visual inspection might overlook.

Security, Privacy, and Broad Clinical Reach

Because clinical neuroimaging involves sensitive patient health information, the platform incorporates advanced AI security and privacy-preserving protocols overseen by Lehigh computer scientist Lichao Sun, PhD, ensuring compliance with medical data governance.

Additional key collaborators on the initiative include Quanzheng Li, PhD, Randy Hirschtick, MD, Daniel Barron, MD, PhD, Boadie Dunlop, MD, and Ying Ding, PhD, providing clinical validation across psychiatric and neurodegenerative patient cohorts, including Alzheimer’s disease, major depression, and related neurological disorders.

“This represents the next generation of AI for scientific discovery,” He noted, “and it could help scientists make discoveries that weren’t possible with earlier tools.”

By bridging the gap between raw medical imaging and rapid, interpretable diagnostic discovery, UniBrain aims to provide an accessible, standardized foundation for precision neurology and translational neuroscience.

Editorial Notes:

• This article was edited by a Neuroscience News editor.
• Journal paper will be reviewed in full upon release.
• Additional context added by our staff.

About this neurotech and schizophrenia Research:

• Media Contact: Katie Kackenmeister
• Source: Lehigh University
• Image Credit: Image credited to Neuroscience News

Why the best learners talk to themselves

49,634 plays|

Tania Lombrozo |

TEDxNewEngland

• October 2025Save talks for later, comment, and more.Create a free accountRead transcript

With more information at our fingertips than ever before, learning should be getting easier … yet the opposite seems to be true, says cognitive scientist Tania Lombrozo. Drawing on her research into how our minds understand and reason, she explores why more information can actually make learning harder — and shares the critical skill all great learners possess.

What if you could read the DNA of your future child?

76,844 plays|

Noor Siddiqui |

TED2026

• April 2026 Save talks for later, comment, and more.Create a free account Read transcript

For all of human history, a baby’s DNA was determined by chance at conception: the so-called genetic lottery. “That era is over,” says Noor Siddiqui, whose company builds screening tools to read the whole genome of embryos in an effort to give parents more genetic information than standard IVF. She makes the case that this kind of technology helps families understand and reduce health risks, raising the question: How much should we know about our children’s DNA before birth? (Followed by a brief Q&A with TED’s vision steward Sal Khan)

North Korean workers treated as ‘slaves’ at Russian drone factory

Julian Ryall10/20/2026October 20, 2026

A new report reveals the plight of North Korean workers at a drone factory in Russia. It also raised concerns that Pyongyang could gain hands-on knowledge of advanced weapons systems through the workers it sends there.

https://p.dw.com/p/5OSRJ

As many as 12,000 North Koreans working at a Russian drone factory are victims of “gross human rights abuses and human trafficking,” according to a new report.

The figures were part of a joint report published on October 14 by the  Washington-based Center for Strategic and International Studies (CSIS) and the Dallas-based George W. Bush Institute. It analyzes how Russia, North Korea, China and Iran are involved in expanding unmanned aerial vehicle (UAV) production at the Yelabuga plant in Tatarstan, and what each country stands to gain.

While Moscow gets access to more than 5,500 drones a month for its war against Ukraine, China benefits from being the main supplier of dual-use components, including batteries, internal electronics, engines and navigation systems.

Iran has supplied Russia with Shahed drones, which has helped Moscow produce upgraded domestic variants, known as Geran.

Meanwhile, North Korea is earning money by retaining as much as 90% of the wages of the laborers, the report stated.

“The geographical locations of these four countries… means that any kind of international interdiction is impossible,” Dan Pinkston, a professor of international relations at the Seoul campus of Troy University, told DW.

North Korea gains critical intel by aiding Russia in Ukraine

Russia and North Korea defy UN resolutions

The joint report says Russia and North Korea are defying United Nations Security Council Resolution 2397, of December 2017, which requires any UN member state to send North Korean workers home within two years.

“The North Korean regime… is selling the labor of its citizens under a state-to-state arrangement, controlling whether they may leave and collecting most of their wages,” it said. This, according to the report, “is the definition of forced labor.”

Song Young-Chae, a Seoul-based academic and human rights activist with the Worldwide Coalition to Stop Genocide in North Korea, says Pyongyang is “effectively conducting slavery of its own people and for its own profit.”

“This is just a terrible situation, and we know that it is not happening only at Yelabuga,” he told DW. “We know there are construction workers in the occupied Ukrainian territories, North Koreans in lumber camps and, of course, tens of thousands of soldiers have been sent to fight — and die — on the front lines in Ukraine,” Song added.

“Life in North Korea is not easy, but these people are being sent to Russia as slaves, with no freedom, and they are only allowed to keep a fraction of the money they earn,” he said.

The challenge, he admits, is to find a way to halt this “trade in humanity.”

How much can realistically be done?

“These countries share a border, so it is easy to transport workers, and no other agency can intervene,” Song said. “Russia and China are both permanent members of the UN Security Council but are ignoring its resolutions on this issue. There is very little that can be done.”

Song’s rights organization is attempting to highlight the plight of North Korean workers and to create a “coalition of like-minded nations and agencies” to protect laborers — although he admits to being “pessimistic.”

Pinkston shares those concerns, as well as fears over the emergence of a powerful bloc that appears to be “ripping up the post-World War II order on the rule of law and human rights” and is intent on taking a confrontational approach to other nations.

The weapons being manufactured at Yelabuga will assist in achieving those aims, he said.

“UN Security Council resolutions on North Korean workers, on weapons proliferation, on the sale of dual-use technologies are all being ignored, so I think we can say that the UN’s sanctions regime is pretty much dead,” Pinkston added.

North Korea developing its drone capabilities

It is clear that both Russia and Ukraine have gained important insights into the modern battlefield through their ongoing conflict, Pinkston said, with the emergence of drone technology arguably one of the most critical advances.

“And like everyone else, North Korea has been watching closely, ‘learning-by-doing’ and innovating its own drone capabilities,” Pinkston said, adding that there are “serious implications” for the South Korean military as it also attempts to keep up its defensive capabilities against a new generation of drones in the North.

Why Russia and North Korea can’t quit each other (for now)

To curb the deployment of North Korean laborers to Russia and the proliferation of weapons, the joint report calls on the US State Department to work with other countries to disrupt human trafficking networks and sanction those responsible.

It also urges the UN to work multilaterally to investigate the Yelabuga complex and to catalogue the abuses that have taken place. Moreover, it demands accountability for those who have carried out abuses and for victims to be protected.

The report adds that foreign companies that are operating as part of the Yelabuga special economic zone — there are “several” European companies and at least one US firm with a presence at the site — should be pressured to leave.

Edited by: Emmy Sasipornkarn

https://www.dw.com/webapi/iframes/widget/en/66496979

Julian Ryall Journalist based in Tokyo, focusing on political, economic and social issues in Japan and Korea

https://p.dw.com/p/5OEyo?at_medium=SocialMedia&at_campaign=Twitter&at_share_source=SharingButton&at_origin=Web

Peter Hille

10/20/2026

Israel’s military has scored major victories against Hamas, Hezbollah and Iran. But around the world, support for the country is slipping. So how isolated is Israel really? And which allies still stand by its side?

https://p.dw.com/p/5OEyo

Israel‘s military power and key alliances remain strong, but its international standing is facing growing challenges.

DW explores how the wars in Gaza, Lebanon and Iran have fueled criticism of Israel, shifting public opinion in Europe and the United States.

While support among younger generations is declining and some governments have imposed sanctions on members of the administration of Prime Minister Benjamin Netanyahu, Israel continues to benefit from close ties with the US, European security partners and several Arab states.

Correction, October 20, 2026: In the graphic shown at 2:37, Norway and Sweden were mistakenly swapped. DW apologizes for the error and any confusion it may have caused.

Peter Hille Peter Hille is a multimedia reporter with a focus on geopolitics.Skip next section Similar stories

Similar stories

Exclusive: Main Street’s budget AI boom

Data: Goldman Sachs. Chart: Erin Davis/Axios Small businesses across America are embracing AI, but usage and spending vary dramatically by state, Axios’ Herb Scribner writes from an exclusive Goldman Sachs survey of 2,474 small business owners. Why it matters: Companies nationwide, big and small, are figuring out how best to incorporate AI into their workforces and what that means for productivity By the numbers: For all the talk about AI’s explosive growth in the enterprise, the survey shows 73% of small businesses are spending $100 or less per month on AI. In Texas, most use AI for marketing copy. In Florida, it’s mainly for productivity.

Trendingnow

1

UnitedHealthcare and Humana are ending Medicare Advantage plans for nearly 1 million people

2

Zuckerberg told Meta directors he might leave California, where he could lose billions to a wealth tax. Meta answered with a new class of stock

3

Martha Stewart says people need to get it through their ‘thick skulls’ that there’s ‘no such thing as balance’—even at 85 she starts work at 7 a.m.

Economy national debt

America’s national debt plan is ‘clearly not in order,’ says Ken Griffin. He believes part of the problem is Europeans who ‘don’t want to go to work’

By

Eleanor Pringle

Senior Reporter, Economics and Markets

Follow

October 21, 2026, 12:21 PM ET

Ken Griffin, CEO of Citadel, at a Citadel Securities conference in New York, Oct. 20, 2026.Michael Nagle—Bloomberg/Getty Images

Citadel CEO Ken Griffin is one of many people growing increasingly concerned about America’s $40 trillion national debt burden, and the budget deficits that add to it year after year. But Griffin has an alternative perspective to some of his peers: This isn’t an issue that can be solved in Washington, D.C., alone, he says.

The feathers of debt hawks have been ruffled further over the past few days following the Treasury’s confirmation that it ran a $2 trillion deficit for fiscal year 2026. More than half of that deficit came from interest rate payments to service the debt.

Economists have a range of ideas on how to improve the U.S. fiscal picture. Some call for policies supporting economic growth to better balance government borrowing versus income. Others say spending needs to be cut to shave budget deficits down as a portion of GDP.

Recommended Video

---

Only 1 Trump gold card visa has been sold so far

Griffin suggests the matter also requires impetus from U.S. allies—namely, them needing to grow their own economies.

Speaking at the Citadel Securities Future of Global Markets event in New York yesterday, Griffin said: “Our fiscal house is clearly not in order.”

When asked where he would begin, the Citadel CEO said, “We really need to play the productivity card as hard as we can.” He added: “In order of what priorities need to be, grow the damn economy.”

But the U.S.’s allies also need to be growing, Griffin added. “You actually want the Europeans to grow their economy,” he explained. “The tariff war with Europe’s a fiasco because at this moment we should use all of our political capital: ‘Europe—you want us to pay for your defense? Guess what? You’re going to go back to work.’

“In France, there are strict limitations on how many hours you can work a week, strict limitations. It’s ridiculous. Like the French want the Americans to provide the defense for Europe, and they don’t want to go to work. This just doesn’t work for the American people.”

France does indeed have labor laws to ensure workers don’t work for more than 48 hours a week, or 44 hours per week on average over a period of 12 consecutive weeks. A break of at least 20 minutes is also legally mandated for every six hours of work, to be taken within or immediately after that period.

The French nation is also famous for a cultural—but not legal—break that many workers take in July or August. Les grandes vacances refers to the phenomenon of workers taking a month out of their five weeks of annual leave to escape busy city hubs and enjoy the countryside. Workers in service industries and essential services often continue working.

However, Griffin’s suggestion that more hours worked will equal more economic growth is perhaps oversimplified. A working-hours study from Charlie Giattino, Esteban Ortiz-Ospina, and Max Roser for Our World in Data showed that in 2023, Norway’s GDP per capita was close to $99,000 for annual hours worked of 1,412. The U.S., by contrast, has GDP per capita of a little under $73,000 for an average of 1,789 hours worked per employee every year.

Griffin’s comments perhaps refer to the debate over defense spending among the G7 countries, a point which President Trump has forcefully made since returning to the White House. The U.S. spends approximately 3.1% to 3.4% of its GDP on defense, with the Pentagon awarded more than $1 trillion in funding earlier this year.

Paid Content

Why smoking remains the U.S.’s ‘invisible’ public health crisis

From PMI U.S.

France’s military budget, per Politico, will be increased to €63.4 billion ($71 billion) for 2027, and tends to be roughly 2% of its GDP.

Further measures

Of course, the U.S. fiscal picture can’t be improved on the actions of allies alone. And Griffin has plenty of other ideas for what to look at next: “We need to make sure that data centers are built in America. But we need more oars in the water; we need more focus on nuclear fusion … We need more focus on robotics. We need more focus on the technologies that will unleash productivity for the next 20 years. Like, that should be the singular focus of our administration from an economic policy perspective.

“Grow our economy and push our allies to engage in meaningful reforms that grow their economies. The entire Western world is in the same boat together … We’re not going to get out of these waters by torturing each other. We’re going to get out of these waters by making the pie bigger for all of us.”

Improving education is also key to unlocking growth, Griffin said, as well as getting immigration policy right: “Every single brilliant person around the world. I want them to live in America.”

Fortune Daily breaks the traditional barrier between audience and newsroom. The show transforms Fortune’s trusted reporting into actionable, conversational, and entertaining insights for an emerging class of business leaders. Watch here.

About the Author

By Eleanor PringleSenior Reporter, Economics and Markets

Follow

Eleanor Pringle is an award-winning senior reporter at Fortune covering news, the economy, and personal finance. Eleanor previously worked as a business correspondent and news editor in regional news in the U.K. She completed her journalism training with the Press Association after earning a degree from the University of East Anglia.

See full bio

Sponsored Stories

These 2026 Glasses Adjust Their Focus Automatically – Near or FarBlueLume Glasses

15 of 2,100

Agents need better memory more than better models

Inbox

The Deep View <newsletter@thedeepview.co> Unsubscribe	Wed, Oct 21, 1:36 PM (19 hours ago)
to me

October 21, 2026   |   Read online Welcome back. New research finds that AI’s use in scientific contexts may be accelerating discovery. Meanwhile, Amazon researchers created an internal tracing system that improves its AI coding agent, Kiro, from its own runs. And Databricks new memory layer for agents could solve one of the most critical problems agents face: A better understanding of a user’s history. —Nat Rubio-Licht IN TODAY’S NEWSLETTER1. Agents need better memory more than better models2. Amazon tests a safer path to self-improving AI3. Can AI free up scientists to think bigger? ENTERPRISE Why persistent memory could transform AI agents What if the path to more capable AI in the real world wasn’t building more powerful models? Databricks launched new technology on Tuesday that offers today’s existing models more extensive context and more persistent knowledge to improve AI performance. The company unveiled MemCurrent, a new memory layer for agents that can retain knowledge across sessions and create institutional memory for enterprises—and it offers some pretty impressive results.Databricks reported these findings on the performance of MemCurrent: Agent success rate in carrying out real-world tasks tripled from 22% to 68%, based on an internal benchmark; and it did things 26% faster and with 31% fewer tokens Using three public benchmarks, MemCurrent performed as well as giving the agent every past conversation, but did it with 12x fewer tokens. Databricks found that MemCurrent outperformed three open-source memory systems, OpenClaw, Mem0, and PropMem, on accuracy and efficiency. This is one of the hardest problems in the AI industry today. It’s why startups like Pathway have built entirely new architectures to fix the endemic memory problems in LLMs. The issue leads to a number of challenges for agents, especially wasting time and money relearning the same behaviors over and over again. That creates a tradeoff to those using agents—you can make them smarter or you can make them faster and cheaper.  Databricks tackles the problem by giving agents two types of memory: episodic (the raw conversation history of what was learned) and derived (what can be discerned from the patterns, preferences, and workflows). Based on Databricks’ own internal benchmark of 114 enterprise tasks, the average task with no memory was completed 22.3% of the time. With only episodic memory, it was completed 35.8% of the time. With only derived memory, it was completed 66.1% of the time. With both episodic and derived memory, it was completed 68.4% of the time. “On your next project, you don’t have to explain the same thing again,” Michael Bendersky, research director of AI at Databricks, told The Deep View. “The agent will … store your preferences and understand how you work on projects. We can propagate these preferences … from project to project implicitly without you doing anything. “Databricks customers can start using MemCurrent today, but keep in mind that it’s an underlying architecture and not a product. It’s now integrated into the company’s AI data assistant, Genie, where it can immediately boost agents. Customers can also use the new Agent Bricks Memory, launching in beta, to add the MemCurrent capabilities to any of their own custom AI agents running any model or harness. It all happens within Databricks’ Lakebase database.Beyond the immediate value for Databricks’ customers, MemCurrent is another sign that many of the most important breakthroughs in AI are coming from better systems surrounding the models and not just the advancements in the models themselves. Memory continues to emerge as one of the most problematic limitations of LLMs, the predominant architecture of today’s AI. The memory problem leads to inconsistencies, slow answers, and above all, inefficiencies that waste a lot of tokens, compute, and money. It’s like having an employee use their time to learn how to do something one day and then forget it at the end of the day and have to waste time learning the same thing over again the next day, and the next day—and so you keep hiring more and more employees to overcompensate. In this analogy, more and more employees are equivalent to the additional GPUs, compute, and data centers required to run agents. Right now, the industry’s solution is to just build more and more data centers, but it’s already very clear that’s not a sustainable or desirable path forward. What we need instead is efficiency breakthroughs like this one from Databricks.