--- title: Intelligence is Free, Now What? Data Systems for, of,... description: As AI inference costs plummet toward near-zero, Berkeley researchers argue data systems must evolve along three axes. First, 'Data Systems For Agents':... canonical: https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Intelligence is Free, Now What? Data Systems for, of, and by Agents – The Berkeley Artificial Intelligence Research Blog | daily.dev og:description: As AI inference costs plummet toward near-zero, Berkeley researchers argue data systems must evolve along three axes. First, 'Data Systems For Agents':... og:url: https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w og:image: https://api.daily.dev/og/posts/UYi71Ae3w.png og:image:alt: Intelligence is Free, Now What? Data Systems for, of, and by Agents – The Berkeley Artificial Intelligence Research Blog og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Intelligence is Free, Now What? Data Systems for, of, and by Agents – The Berkeley Artificial Intelligence Research Blog **[BAIR](https://daily.dev/sources/bair)** · 14 min read · 1 upvotes · 0 comments ## Summary As AI inference costs plummet toward near-zero, Berkeley researchers argue data systems must evolve along three axes. First, 'Data Systems For Agents': databases must be redesigned to handle thousands of concurrent agentic queries, exploiting multi-query optimization, approximate query processing, and proactive steering rather than passive execution. Second, 'Data Systems Of Agents': a new substrate is needed for agent swarms to manage shared state, structured memory (organized across multiple attributes for precise retrieval), concurrent edits, coordination, and failure recovery—going well beyond markdown files and knowledge graphs. Third, 'Data Systems By Agents': near-free intelligence enables agents to synthesize entire custom data systems (OLAP engines, key-value stores) from scratch in minutes, but verifying correctness against incomplete specifications remains the core challenge, with verification agents and formal proof generation as promising directions. The post concludes that agents and data systems will co-evolve, with boundaries blurring as agents design the very systems they run on. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why do multiple AI agents querying a database produce so much duplicate work? On text-to-SQL benchmarks where multiple agents attempt the same task, only 10-20% of the generated sub-plans are actually distinct, meaning 80-90% of sub-queries perform duplicate work. Despite this redundancy, task success rates increase with more agentic attempts, so the duplication helps overall accuracy even though it wastes data system resources. _Anyone optimizing agent-driven query workloads can follow this research thread on daily.dev._ ### Why are unstructured markdown files not enough for long-term AI agent memory at scale? Markdown-file-based memory breaks down because retrieving all potentially relevant fragments and stuffing them into a limited context window becomes infeasible as agent workloads grow, and even with larger context windows there are latency costs to including irrelevant data. Structured memory instead organizes recall across attributes like tables, columns, and operation type so agents retrieve only pertinent, corrective information. _Developers designing agent memory systems can track this structured-memory approach on daily.dev._ ### How can AI agents build a complete custom database engine from scratch instead of using a general-purpose system? Agentic pipelines such as Bespoke OLAP and GenDB can synthesize a complete, workload-specific analytical engine in minutes to a few hours for a cost of just a few dollars, and similarly custom key-value stores can be generated targeted to a specific workload. These engines are treated as disposable and regenerated whenever the workload shifts, though verifying correctness against imperfect specifications remains a challenge. _Engineers evaluating agent-generated infrastructure can follow this workload-synthesis approach on daily.dev._ ## Similar posts on daily.dev - [The era of agentic chaos and how data will save us](https://daily.dev/posts/the-era-of-agentic-chaos-and-how-data-will-save-us-sn17lztqw) · MIT Technology Review · 1 upvotes · 0 comments - [Designing the agent-ready data stack](https://daily.dev/posts/designing-the-agent-ready-data-stack-hb3dcwedk) · InfoWorld · 0 upvotes · 0 comments - [AI Infrastructure Starts With Data: Why Databases Are Critical in the Agentic AI Era](https://daily.dev/posts/ai-infrastructure-starts-with-data-why-databases-are-critical-in-the-agentic-ai-era-fv6einegp) · BigData Boutique blog · 2 upvotes · 0 comments - [Data-Native AI Agents: Why Agents Must Move to Your Data](https://daily.dev/posts/data-native-ai-agents-why-agents-must-move-to-your-data-xuwcq4btu) · databricks · 1 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#ai-inference](https://daily.dev/tags/ai-inference) [View this post on daily.dev](https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Intelligence is Free, Now What? Data Systems for, of, and by Agents – The Berkeley Artificial Intelligence Research Blog","url":"https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w"},"datePublished":"2026-07-07T17:06:54.172Z","dateModified":"2026-09-14T07:24:55.146Z","description":"As AI inference costs plummet toward near-zero, Berkeley researchers argue data systems must evolve along three axes. First, 'Data Systems For Agents':...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2a959bba780915425ae33f321e521641?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/2a959bba780915425ae33f321e521641?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"BAIR","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"BAIR","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/bair","url":"https://daily.dev/sources/bair"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,ai-inference","timeRequired":"PT14M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"BAIR","item":"https://daily.dev/sources/bair"},{"@type":"ListItem","position":3,"name":"Intelligence is Free, Now What? Data Systems for, of, and by Agents – The Berkeley Artificial Intelligence Research Blog"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/intelligence-is-free-now-what-data-systems-for-of-and-by-agents-the-berkeley-artificial-intell-uyi71ae3w#faq","mainEntity":[{"@type":"Question","name":"Why do multiple AI agents querying a database produce so much duplicate work?","acceptedAnswer":{"@type":"Answer","text":"On text-to-SQL benchmarks where multiple agents attempt the same task, only 10-20% of the generated sub-plans are actually distinct, meaning 80-90% of sub-queries perform duplicate work. Despite this redundancy, task success rates increase with more agentic attempts, so the duplication helps overall accuracy even though it wastes data system resources. Anyone optimizing agent-driven query workloads can follow this research thread on daily.dev."}},{"@type":"Question","name":"Why are unstructured markdown files not enough for long-term AI agent memory at scale?","acceptedAnswer":{"@type":"Answer","text":"Markdown-file-based memory breaks down because retrieving all potentially relevant fragments and stuffing them into a limited context window becomes infeasible as agent workloads grow, and even with larger context windows there are latency costs to including irrelevant data. Structured memory instead organizes recall across attributes like tables, columns, and operation type so agents retrieve only pertinent, corrective information. Developers designing agent memory systems can track this structured-memory approach on daily.dev."}},{"@type":"Question","name":"How can AI agents build a complete custom database engine from scratch instead of using a general-purpose system?","acceptedAnswer":{"@type":"Answer","text":"Agentic pipelines such as Bespoke OLAP and GenDB can synthesize a complete, workload-specific analytical engine in minutes to a few hours for a cost of just a few dollars, and similarly custom key-value stores can be generated targeted to a specific workload. These engines are treated as disposable and regenerated whenever the workload shifts, though verifying correctness against imperfect specifications remains a challenge. Engineers evaluating agent-generated infrastructure can follow this workload-synthesis approach on daily.dev."}}]} ```