pith. machine review for the scientific record. sign in

arxiv: 2604.14590 · v2 · submitted 2026-04-16 · 💻 cs.DC · cs.AI

Recognition: unknown

AgileLog: A Forkable Shared Log for Agents on Data Streams

Shreesha G. Bhat , Tony Hong , Michael Noguera , Ramnatthan Alagappan , Aishwarya Ganesan
Authors on Pith no claims yet

Pith reviewed 2026-05-10 10:15 UTC · model grok-4.3

classification 💻 cs.DC cs.AI
keywords shared logdata streamsAI agentsforkingperformance isolationstreaming systemsBolt
0
0 comments X

The pith

Shared logs for streaming data must support forks so AI agents can operate without interfering with each other or the main stream.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that AI agents, which use large language models to carry out natural-language tasks over live data streams, expose limitations in today's streaming systems. These systems cannot prevent one agent's work from slowing others or handle agent writes safely. To fix this, the authors propose that the shared log abstraction at the heart of streaming data needs explicit support for creating forks. AgileLog adds forking primitives that let agents branch the log, and its Bolt implementation makes those branches cheap while delivering both logical and performance isolation. If the argument holds, agents could run concurrent high-level reasoning tasks on the same data feed without the conflicts that currently arise.

Core claim

AgileLog is a shared log abstraction that supplies forking primitives so agents can create independent branches of a data stream. The Bolt implementation realizes these primitives with techniques that keep fork creation inexpensive and enforce isolation between branches and the original log.

What carries the argument

The fork primitive on the shared log, which produces a new, isolated branch that agents can use for private reads and writes.

If this is right

  • Agents can each maintain a private fork of the stream for independent reasoning and writes.
  • Performance of one agent's task no longer slows the main stream or other agents.
  • Writes from agents stay confined to their forks and do not corrupt the shared log.
  • Traditional programs and agents can coexist on the same stream with separation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Multi-agent systems could let each agent explore a different hypothesis by forking the same live feed and later comparing results.
  • The forking idea might extend to collaborative human-AI workflows where participants branch data views without mutual disruption.
  • Developers could treat forks as lightweight workspaces for testing agent behaviors before merging changes back to the primary stream.

Load-bearing premise

Current streaming systems have no way to stop performance interference or manage agent writes safely, and adding forking will solve both problems without new drawbacks.

What would settle it

A measurement showing that forks in the Bolt implementation either cost high overhead or allow one agent's activity to degrade the performance of the main log or other forks would disprove the proposal.

Figures

Figures reproduced from arXiv: 2604.14590 by Aishwarya Ganesan, Michael Noguera, Ramnatthan Alagappan, Shreesha G. Bhat, Tony Hong.

Figure 1
Figure 1. Figure 1: AgileLog Interface. fraud events. However, these events must remain private to the child. cFork provides this required isolation: writes on children forks remain private to them and consumers of the parent will not see them. Thus, cForks enable a unique form of isolation that we call unidirectional write isolation: writes on the parent are visible to the children, but not the other way around. This is in c… view at source ↗
Figure 2
Figure 2. Figure 2: (a) and (b) show cFork semantics. In 2(a), the cFork acts as a live read-only copy of the parent. In 2(b), in addition, records are inserted in the live copy (e.g., Z on green fork) that are linearizably interleaved with the parent records. Severed Forks. AgileLog also offers regular forks via a sFork call, where the child is “severed” from the parent after the fork point. By default, a severed fork is cre… view at source ↗
Figure 3
Figure 3. Figure 3: Diskless Primer. The figure shows how appends and reads work (for entries in log L). Only L’s metadata is shown in the metadata layer. data layer of the object identifier (O) and the metadata for each record within O (i.e., the log identifier and byte ranges within O) (a4). The metadata layer then sequences these records by logging this metadata to its consensus log and executing them as SMR operations. Sp… view at source ↗
Figure 4
Figure 4. Figure 4: Continuous Inheritance. G is cForked to get R (blue line is the fork point). (a) and (b) show how BoltNaiveCF and Bolt implement cForks, respectively. Lookup of position 2 in R is also shown (red boxes). metadata updates to the child’s metadata. This metadata prop￾agation indicates the presence of new records on the parent to the child. Since operations on the child also flow through the metadata layer, th… view at source ↗
Figure 6
Figure 6. Figure 6: Parent Throughput during Fork Creation. 0 0.5 1 1.5 2 2.5 3 Kafka Bolt E2E Latency Norm. (ms)w/o analytics w/ analytics (a) Isolation from Reads 0 5 10 15 20 25 30 35 avg p99 E2E Latency (ms) w/o cFork w/ cFork (b) Bolt: Isolation from Appends [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Performance Impact on Parent with many cForks. with Bolt, the parent’s performance remains unaffected. 6.2 Performance Isolation We next show that Bolt provides performance isolation for a latency-critical (lc) workload from a read-heavy workload resembling agentic data analysis. The lc-workload appends and reads records at the tail. The analysis workload reads records in bulk from the same log (up to an o… view at source ↗
Figure 10
Figure 10. Figure 10: Metadata Lookup vs. cFork Depth. 10−1 100 101 102 103 104 105 106 10−1 100 101 102 103 10 Promote Latency (us) 4 #Promoted Records Bolt DC (Kafka) [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 13
Figure 13. Figure 13: Real-time E2E La￾tency w/ Testing Agent. 0 5 10 15 20 25 30 35 40 45 50 55 60 Throughput (KOps) Time (s) Kafka Append 0 5 10 15 20 25 25 30 35 40 45 50 55 60 Throughput (KOps) Time (s) Bolt Fork Promote Squash [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Supply-Chain Agent: Consumer Throughput. one run of the agent and replay the trace on both systems [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
read the original abstract

In modern data-streaming systems, alongside traditional programs, a new type of entity has emerged that can interact with streaming data: AI agents. Unlike traditional programs, AI agents use LLM reasoning to accomplish high-level tasks specified in natural language over streaming data. Unfortunately, current streaming systems cannot fully support agents: they lack the fundamental mechanisms to avoid the performance interference caused by agentic tasks and to safely handle agentic writes. We argue that the shared log, the core abstraction underlying streaming data, must support creating forks of itself, and that such a forkable shared log serves as a great substrate for agents acting on streaming data. We propose AgileLog, a new shared log abstraction that provides novel forking primitives for agentic use cases. We design Bolt, an implementation of the AgileLog abstraction, that uses novel techniques to make forks cheap, and provide logical and performance isolation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper claims that current data-streaming systems cannot fully support AI agents because they lack mechanisms to avoid performance interference from agentic tasks and to safely handle agentic writes. It argues that the shared log abstraction must support forking, proposes AgileLog as a new shared log abstraction with novel forking primitives, and presents Bolt as an implementation that uses novel techniques to make forks cheap while providing logical and performance isolation for agentic use cases.

Significance. If the proposed forking primitives and their implementation in Bolt deliver the claimed isolation without significant overhead or consistency issues, this work could be significant for the field of distributed systems and data streaming by providing a substrate that accommodates both traditional programs and LLM-based agents. It addresses an emerging need as AI agents become more prevalent in interacting with streaming data. However, the current manuscript is primarily a design proposal with high-level claims, and its significance depends on future validation through detailed mechanisms, proofs, or experiments.

major comments (3)
  1. [Abstract] Abstract: The claim that 'current streaming systems cannot fully support agents' due to performance interference and unsafe writes is not substantiated with specific examples or comparisons to systems like Apache Kafka or Apache Flink; this assumption is load-bearing for motivating AgileLog.
  2. [Design of AgileLog and Bolt] Design of AgileLog and Bolt: The description of forking primitives lacks details on fork semantics (e.g., whether forks are copy-on-write, how concurrent writes are handled, or conflict resolution for agentic writes), which is necessary to evaluate the claim of logical and performance isolation without reconciliation overhead.
  3. [Evaluation] Evaluation section: No experimental results, benchmarks, or formal analysis are provided to support the claims that forks are 'cheap' and provide isolation; this undermines the practical significance of the proposal.
minor comments (1)
  1. [Introduction] Introduction: Some terminology like 'agentic tasks' and 'agentic writes' could be defined more clearly for readers unfamiliar with AI agent literature.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We clarify that the work is a design proposal for a new shared-log abstraction and its implementation, and we address each major comment below with planned revisions where the manuscript can be strengthened.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that 'current streaming systems cannot fully support agents' due to performance interference and unsafe writes is not substantiated with specific examples or comparisons to systems like Apache Kafka or Apache Flink; this assumption is load-bearing for motivating AgileLog.

    Authors: We agree that the abstract would be strengthened by concrete examples. The full manuscript motivates the claim in the introduction by noting how agentic workloads can induce latency variance and resource contention in systems such as Kafka (e.g., when long-running LLM tasks share partitions with latency-sensitive consumers) and Flink (e.g., when speculative writes from agents violate exactly-once guarantees). We will revise the abstract to include one or two brief, specific illustrations of these issues while preserving the high-level motivation. revision: partial

  2. Referee: [Design of AgileLog and Bolt] Design of AgileLog and Bolt: The description of forking primitives lacks details on fork semantics (e.g., whether forks are copy-on-write, how concurrent writes are handled, or conflict resolution for agentic writes), which is necessary to evaluate the claim of logical and performance isolation without reconciliation overhead.

    Authors: The current description emphasizes the abstraction-level benefits. We will expand the design section to explicitly state that forks are copy-on-write, that concurrent writes are routed to isolated branch logs to achieve performance isolation, and that agentic writes remain confined to their fork without requiring cross-branch reconciliation or conflict resolution. These additions will make the isolation claims easier to evaluate. revision: yes

  3. Referee: [Evaluation] Evaluation section: No experimental results, benchmarks, or formal analysis are provided to support the claims that forks are 'cheap' and provide isolation; this undermines the practical significance of the proposal.

    Authors: We acknowledge that the manuscript contains no empirical benchmarks or formal proofs, as it focuses on the abstraction and implementation architecture. We will add an analysis subsection that derives the expected cost of fork creation from the copy-on-write and branch-isolation techniques and will include preliminary micro-benchmark sketches based on the Bolt design. A comprehensive experimental evaluation remains future work. revision: partial

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's core argument identifies limitations in existing streaming systems regarding agentic workloads and proposes AgileLog as a new forkable shared-log abstraction with forking primitives implemented in Bolt for cheap forks and isolation. No load-bearing step reduces by construction to its own inputs: there are no equations, fitted parameters renamed as predictions, self-definitional claims, or uniqueness theorems imported via self-citation that force the result. The proposal rests on design rationale and use-case motivation rather than circular derivation, making the central claims self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The central claim rests on the new abstractions AgileLog and Bolt plus the assumption that forking solves interference and safety issues. No free parameters are fitted to data. No invented physical entities; the new entities are software abstractions.

axioms (2)
  • domain assumption Shared logs are the core abstraction in streaming data systems.
    Stated in the abstract as the foundation for the argument.
  • domain assumption AI agents require mechanisms to avoid performance interference and safely handle writes.
    Presented as the motivation for needing forks.
invented entities (2)
  • AgileLog no independent evidence
    purpose: Forkable shared log abstraction with novel forking primitives for agents.
    New abstraction proposed to address agentic use cases.
  • Bolt no independent evidence
    purpose: Implementation of AgileLog using novel techniques for cheap forks and isolation.
    New implementation claimed to realize the abstraction.

pith-pipeline@v0.9.0 · 5465 in / 1440 out tokens · 29700 ms · 2026-05-10T10:15:57.986163+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

142 extracted references · 13 canonical work pages · 4 internal anchors

  1. [1]

    https://github.com/tuannvm/kafka-mcp- server

    A Model Context Protocol (MCP) server for Apache Kafka. https://github.com/tuannvm/kafka-mcp- server

  2. [2]

    https://flink.apache.org/

    Apache Flink: Stateful Computations over Data Streams. https://flink.apache.org/. Apache Soft- ware Foundation

  3. [3]

    https://pulsar

    Apache Pulsar Functions Overview. https://pulsar. apache.org/docs/next/functions-overview/. Apache Software Foundation

  4. [4]

    https://supabase.com/docs/guides/deployment/ branching

    Branching in Supabase to Test and Preview Changes. https://supabase.com/docs/guides/deployment/ branching

  5. [5]

    https: //neon.com/docs/guides/branching-test-queries

    Branching Workflows to Test Queries in Neon. https: //neon.com/docs/guides/branching-test-queries

  6. [6]

    https: //neon.com/branching/branching-for-agents

    Build Versioning / Checkpoints for your Agent. https: //neon.com/branching/branching-for-agents

  7. [7]

    https://docs.cloudlab.us/ hardware.html

    Cloudlab Hardware. https://docs.cloudlab.us/ hardware.html

  8. [8]

    https://en.wikipedia.org/wiki/ Confluent

    Confluent, Inc. https://en.wikipedia.org/wiki/ Confluent. Accessed: 2025-08-28

    2025

  9. [9]

    Dolt is Git for Data! https://github.com/dolthub/ dolt

  10. [10]

    https: //www.tigerdata.com/blog/fluid-storage- forkable-ephemeral-durable-infrastructure- age-of-agents

    Fluid Storage: Forkable, Ephemeral, Durable Infrastructure for the Age of Agents. https: //www.tigerdata.com/blog/fluid-storage- forkable-ephemeral-durable-infrastructure- age-of-agents

  11. [11]

    https://www.tigrisdata.com/blog/fork- buckets-like-code/

    Fork Buckets Like You Fork Code | Tigris Object Storage. https://www.tigrisdata.com/blog/fork- buckets-like-code/

  12. [12]

    com/concepts/dolthub/forks

    Forks of a Database with Dolt.https://docs.dolthub. com/concepts/dolthub/forks

  13. [13]

    https://streamnative.io/blog/introducing-the- streamnative-mcp-server-connecting-streaming- data-to-ai-agents

    Introducing the StreamNative MCP Server: Connecting Streaming Data to AI Agents. https://streamnative.io/blog/introducing-the- streamnative-mcp-server-connecting-streaming- data-to-ai-agents

  14. [14]

    https://lenses.io/

    Lenses IO. https://lenses.io/

  15. [15]

    https://neon.com/blog/marble-is-shipping- faster-with-neon-branching

    Marble is Shipping Faster with Neon Branch- ing. https://neon.com/blog/marble-is-shipping- faster-with-neon-branching

  16. [16]

    https://netflixtechblog.com/streaming- sql-in-data-mesh-0d83f5a00d08

    Netflix Data Mesh: Streaming SQL in Data Mesh. https://netflixtechblog.com/streaming- sql-in-data-mesh-0d83f5a00d08

  17. [17]

    https://developer.nvidia.com/blog/building- ai-agents-to-automate-software-test-case- creation/

    NVIDIA Hephaestus: Building AI Agents to Automate Software Test Case Creation. https://developer.nvidia.com/blog/building- ai-agents-to-automate-software-test-case- creation/

  18. [18]

    https://www.linkedin.com/blog/engineering/ infrastructure/production-testing-with-dark- canaries

    Production Testing with Dark Canaries. https://www.linkedin.com/blog/engineering/ infrastructure/production-testing-with-dark- canaries

  19. [19]

    https://streamnative.io/

    StreamNative, Inc. https://streamnative.io/

  20. [20]

    https://github.com/nodece/ pulsar-mcp-server, 2025

    pulsar-mcp-server: A Model Context Protocol Server for Apache Pulsar. https://github.com/nodece/ pulsar-mcp-server, 2025

    2025

  21. [21]

    Olive: Distributed Point-in-Time Branch- ing Storage for Real Systems

    Marcos Kawazoe Aguilera, Susan Spence, and Alis- tair C Veitch. Olive: Distributed Point-in-Time Branch- ing Storage for Real Systems. In Proceedings of the 3rd Symposium on Networked Systems Design and Im- plementation (NSDI ’06), San Jose, CA, May 2006

    2006

  22. [22]

    Aiven Inkless: Diskless Topics for Apache Kafka

    Aiven. Aiven Inkless: Diskless Topics for Apache Kafka. https://aiven.io/inkless

  23. [23]

    Automated Unit Test improvement using Large Language Models at Meta

    Nadia Alshahwan, Jubin Chheda, Anastasia Finogen- ova, Beliz Gokkaya, Mark Harman, Inna Harper, Alexandru Marginean, Shubho Sengupta, and Eddy Wang. Automated Unit Test improvement using Large Language Models at Meta. In Companion Proceed- ings of the 32nd ACM International Conference on the Foundations of Software Engineering, 2024. 13

    2024

  24. [24]

    CXLfork: Fast Remote Fork over CXL Fabrics

    Chloe Alverti, Stratos Psomadakis, Burak Ocalan, Shashwat Jaiswal, Tianyin Xu, and Josep Torrellas. CXLfork: Fast Remote Fork over CXL Fabrics. InPro- ceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2, ASPLOS ’25, 2025

    2025

  25. [25]

    Best practices design patterns: optimizing Amazon S3 perfor- mance

    Amazon Web Services, Inc. Best practices design patterns: optimizing Amazon S3 perfor- mance. https://docs.aws.amazon.com/AmazonS3/ latest/userguide/optimizing-performance.html

  26. [26]

    Model Context Protocol Specification

    Anthropic and Community Contributors. Model Context Protocol Specification. https: //modelcontextprotocol.io/specification

  27. [27]

    Apache. Kakfa. http://kafka.apache.org/

  28. [28]

    Apache. Pulsar. https://pulsar.apache.org/

  29. [29]

    Kafka mirroring (mir- rormaker)

    Apache Software Foundation. Kafka mirroring (mir- rormaker). https://cwiki.apache.org/confluence/ pages/viewpage.action?pageId=27846330, 2017

    2017

  30. [30]

    Amazon Kinesis

    AWS. Amazon Kinesis. https://aws.amazon.com/ kinesis/

  31. [31]

    Build- ing an Event-Driven Agentic AI System with Apache Kafka on Confluent Cloud and watsonx Orches- trate

    Ahmed Azraq and Moises Dominguez Garcia. Build- ing an Event-Driven Agentic AI System with Apache Kafka on Confluent Cloud and watsonx Orches- trate. https://developer.ibm.com/tutorials/event- driven-agentic-ai-system-confluent-watsonx- orchestrate/, 2026

    2026

  32. [32]

    Transparent, Evaluable, and Acces- sible Data Agents: A Proof-of-Concept Framework

    Nooshin Bahador. Transparent, Evaluable, and Acces- sible Data Agents: A Proof-of-Concept Framework. arXiv preprint arXiv:2509.24127, 2025

    work page arXiv 2025

  33. [33]

    How to detect fraudulent clicks in a real-time ad system

    Gourav Singh Bais. How to detect fraudulent clicks in a real-time ad system. https://www.redpanda.com/ blog/detect-fraudulent-clicks-real-time-ads

  34. [34]

    Vir- tual Consensus in Delos

    Mahesh Balakrishnan, Jason Flinn, Chen Shen, Mihir Dharamshi, Ahmed Jafri, Xiao Shi, Santosh Ghosh, Hazem Hassan, Aaryaman Sagar, Rhed Shi, et al. Vir- tual Consensus in Delos. In Proceedings of the 14th USENIX Conference on Operating Systems Design and Implementation (OSDI ’20), Banff, Canada, November 2020

    2020

  35. [35]

    Mahesh Balakrishnan, Dahlia Malkhi, Vijayan Prab- hakaran, Ted Wobber, Michael Wei, and John D. Davis. CORFU: A Shared Log Design for Flash Clusters. In Proceedings of the 9th Symposium on Networked Sys- tems Design and Implementation (NSDI ’12), San Jose, CA, April 2012

    2012

  36. [36]

    Bhat, Tony Hong, Xuhao Luo, Jiyu Hu, Aishwarya Ganesan, and Ramnatthan Alagappan

    Shreesha G. Bhat, Tony Hong, Xuhao Luo, Jiyu Hu, Aishwarya Ganesan, and Ramnatthan Alagappan. Low End-to-End Latency atop a Speculative Shared Log with Fix-Ante Ordering. In Proceedings of the 19th USENIX Conference on Operating Systems Design and Implementation (OSDI ’25), Boston, MA, July 2025

    2025

  37. [37]

    How Often Should Fraud Detection Rules be Reviewed to Maintain Accuracy? https://www.caf.io/post/how-often-should- fraud-detection-rules-be-reviewed-to-maintain- accuracy

    Caf.io. How Often Should Fraud Detection Rules be Reviewed to Maintain Accuracy? https://www.caf.io/post/how-often-should- fraud-detection-rules-be-reviewed-to-maintain- accuracy

  38. [38]

    AI-Powered Data Exploration on SQL, and NoSQL DBs

    Calliope. AI-Powered Data Exploration on SQL, and NoSQL DBs. https://calliope.ai/

  39. [39]

    Arda Celik and Qusay H. Mahmoud. A review of large language models for automated test case generation. Machine Learning and Knowledge Extraction , 7(3), 2025

    2025

  40. [40]

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Mar- cel Blistein, Ori Ram, Dan Zhang, Evan Rosen, et al. Gemini 2.5: Pushing the frontier with advanced reason- ing, multimodality, long context, and next generation agentic capabilities. arXiv preprint arXiv:2507.06261, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [41]

    Confluent Streaming Agents

    Confluent. Confluent Streaming Agents. https://www. confluent.io/product/streaming-agents/

  42. [42]

    The Confluent Real-Time Context En- gine: Power AI Agents and Apps With Instant Intel- ligence

    Confluent, Inc. The Confluent Real-Time Context En- gine: Power AI Agents and Apps With Instant Intel- ligence. https://www.confluent.io/press-release/ real-time-context-engine/ , 2025

    2025

  43. [43]

    Streaming agents examples and tutorials with confluent cloud

    Confluent, Inc. Streaming agents examples and tutorials with confluent cloud. https: //docs.confluent.io/cloud/current/ai/streaming- agents/streaming-agents-examples.html# financial-fraud-detection, 2026

    2026

  44. [44]

    Streaming agents examples with confluent cloud — customer support agent

    Confluent, Inc. Streaming agents examples with confluent cloud — customer support agent. https://docs.confluent.io/cloud/ current/ai/streaming-agents/streaming-agents- examples.html#customer-support-agent, 2026

    2026

  45. [45]

    Streaming Agents Examples — Step 1: Create Content Analysis Agents

    Confluent, Inc. Streaming Agents Examples — Step 1: Create Content Analysis Agents. https: //docs.confluent.io/cloud/current/ai/streaming- agents/streaming-agents-examples.html#step-1- create-content-analysis-agents , 2026

    2026

  46. [46]

    Tardis: A branch-and-merge approach to weak consistency

    Natacha Crooks, Youer Pu, Nancy Estrada, Trinabh Gupta, Lorenzo Alvisi, and Allen Clement. Tardis: A branch-and-merge approach to weak consistency. In 14 Proceedings of the 2016 International Conference on Management of Data, SIGMOD ’16, page 1615–1628, New York, NY , USA, 2016. Association for Computing Machinery

    2016

  47. [47]

    New in Confluent Cloud: Tableflow, Freight Clusters, Apache Flink AI Enhancements, and More

    Yashwanth Dasari. New in Confluent Cloud: Tableflow, Freight Clusters, Apache Flink AI Enhancements, and More. https://www.confluent.io/blog/2025-q1- confluent-cloud-launch/#freight-clusters-q1 , 2025

    2025

  48. [48]

    Native Search and Retrieval on Databases with Tiger Data

    Tiger Data. Native Search and Retrieval on Databases with Tiger Data. https://www.tigerdata.com/blog/ postgres-for-agents

  49. [49]

    State of AI Agents 2026

    Databricks. State of AI Agents 2026. https: //www.databricks.com/sites/default/files/2026- 01/State-of-AI-Agents-2026-Final.pdf , 2026

    2026

  50. [50]

    Two Paths to Context: When GenAI Agents Need a Real-Time Context En- gine like DeltaStream — and When They Don’t

    DeltaStream, Inc. Two Paths to Context: When GenAI Agents Need a Real-Time Context En- gine like DeltaStream — and When They Don’t. https://www.deltastream.io/blog/two-paths-to- context-when-genai-agents-need-a-real-time- context-engine-like-deltastream-and-when-they- dont/, 2025

    2025

  51. [51]

    Scalog: Seamless Reconfiguration and Total Order in a Scalable Shared Log

    Cong Ding, David Chu, Evan Zhao, Xiang Li, Lorenzo Alvisi, and Robbert Van Renesse. Scalog: Seamless Reconfiguration and Total Order in a Scalable Shared Log. In Proceedings of the 17th Symposium on Net- worked Systems Design and Implementation (NSDI ’20), Santa Clara, CA, February 2020

    2020

  52. [52]

    The design and operation of CloudLab

    Dmitry Duplyakin, Robert Ricci, Aleksander Mar- icq, Gary Wong, Jonathon Duerig, Eric Eide, Leigh Stoller, Mike Hibler, David Johnson, Kirk Webb, Aditya Akella, Kuangching Wang, Glenn Ricart, Larry Landweber, Chip Elliott, Michael Zink, Emmanuel Cecchet, Snigdhaswin Kar, and Prabodh Mishra. The design and operation of CloudLab. In Proceedings of the USENI...

    2019

  53. [53]

    uReplicator: Improvement of apache kafka mirrormaker

    Uber Engineering. uReplicator: Improvement of apache kafka mirrormaker. https://github.com/ uber/uReplicator, 2016

    2016

  54. [54]

    Fabi: AI Data Analysts

    Fabi. Fabi: AI Data Analysts. https://www.fabi.ai/ product/analyst-agent

  55. [55]

    Introducing Real-Time Context Engine: Simplified Context Engineering With Real-Time, Pro- cessed Data for AI

    Sean Falconer. Introducing Real-Time Context Engine: Simplified Context Engineering With Real-Time, Pro- cessed Data for AI. https://www.confluent.io/blog/ introducing-real-time-context-engine-ai/

  56. [56]

    AI Agents in Supply Chain

    Matthew Finio and Amanda Downie. AI Agents in Supply Chain. https://www.ibm.com/think/topics/ ai-agents-supply-chain , 2024

    2024

  57. [57]

    How to Test Safely in Production with Feature Flags? https://www.flagsmith.com/blog/ testing-in-production

    Flagsmith. How to Test Safely in Production with Feature Flags? https://www.flagsmith.com/blog/ testing-in-production

  58. [58]

    FlexLog: A Shared Log for Stateful Serverless Computing

    Dimitra Giantsidi, Emmanouil Giortamis, Nathaniel Tornow, Florin Dinu, and Pramod Bhatotia. FlexLog: A Shared Log for Stateful Serverless Computing. In Proceedings of the 32nd International Symposium on High-Performance Parallel and Distributed Comput- ing, pages 195–209, 2023

    2023

  59. [59]

    Ioana Giurgiu and Michael E. Nidd. Supporting Dy- namic Agentic Workloads: How Data and Agents In- teract. arXiv preprint arXiv:2512.09548, 2025

    work page arXiv 2025

  60. [60]

    Building LinkedIn’s Real-time Activity Data Pipeline

    Ken Goodhope, Joel Koshy, Jay Kreps, Neha Narkhede, Richard Park, Jun Rao, and Victor Yang Ye. Building LinkedIn’s Real-time Activity Data Pipeline. IEEE Data Eng. Bull., 35(2):33–45, 2012

    2012

  61. [61]

    Google. Pub/Sub. https://cloud.google.com/ pubsub

  62. [62]

    Agents meet databases: The future of agentic architectures

    Thibaut Gourdel. Agents meet databases: The future of agentic architectures. https: //thenewstack.io/agents-meet-databases-the- future-of-agentic-architectures/ , 2025

    2025

  63. [63]

    2026 Confluent Predictions Report

    Confluent Technology Strategy Group. 2026 Confluent Predictions Report. https://assets.confluent.io/ m/62a1ef550db9e6a9/original/2026-Predictions- Program-DPF-EB.pdf, 2026

    2026

  64. [64]

    Confluent’s Real-Time Agents Build on Kafka Streaming Data

    Jelani Harper. Confluent’s Real-Time Agents Build on Kafka Streaming Data. https://thenewstack.io/confluents-real-time- agents-build-on-kafka-streaming-data/ , 2025

    2025

  65. [65]

    Measuring Massive Multitask Language Understanding

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Stein- hardt. Measuring Massive Multitask Language Under- standing. arXiv preprint arXiv:2009.03300, 2020

    work page internal anchor Pith review arXiv 2009

  66. [66]

    Random- ized dynamic graph algorithms with polylogarithmic time per operation

    Monika Rauch Henzinger and Valerie King. Random- ized dynamic graph algorithms with polylogarithmic time per operation. In Proceedings of the Twenty- Seventh Annual ACM Symposium on Theory of Com- puting, STOC ’95, page 519–527, New York, NY , USA,

  67. [68]

    Herlihy and Jeannette M

    Maurice P. Herlihy and Jeannette M. Wing. Lineariz- ability: A Correctness Condition for Concurrent Ob- jects. ACM Trans. Program. Lang. Syst., 12(3), July 1990

    1990

  68. [69]

    Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions

    Xinyi Hou, Yanjie Zhao, Shenao Wang, and Haoyu Wang. Model Context Protocol (MCP): Landscape, Se- curity Threats, and Future Research Directions. arXiv preprint arXiv:2503.23278, 2025. 15

    work page internal anchor Pith review arXiv 2025

  69. [70]

    How we continuously update our fraud detectors

    Inscribe.ai. How we continuously update our fraud detectors. https://www.inscribe.ai/blog/how-we- continuously-update-our-fraud-detectors

  70. [71]

    Boki: Stateful Serverless Computing with Shared Logs

    Zhipeng Jia and Emmett Witchel. Boki: Stateful Serverless Computing with Shared Logs. In Proceed- ings of the 28th ACM Symposium on Operating Sys- tems Principles (SOSP ’21), Virtual, October 2021

    2021

  71. [72]

    Unleash real-time agen- tic ai with streaming agents on confluent cloud and couchbase

    Jared Jones and Yves Laurent. Unleash real-time agen- tic ai with streaming agents on confluent cloud and couchbase. https://www.couchbase.com/blog/real- time-agentic-ai-confluent-cloud

  72. [73]

    KIP-1150: Diskless Topics for Kafka

    Kafka. KIP-1150: Diskless Topics for Kafka. https://cwiki.apache.org/confluence/display/ KAFKA/KIP-1150%3A+Diskless+Topics

  73. [74]

    Datacenter RPCs can be general and fast

    Anuj Kalia, Michael Kaminsky, and David Andersen. Datacenter RPCs can be general and fast. In Proceed- ings of the 16th Symposium on Networked Systems Design and Implementation (NSDI ’19), Boston, MA, February 2019

    2019

  74. [75]

    Kafka, samza and the unix philosophy of distributed data

    Martin Kleppmann and Jay Kreps. Kafka, samza and the unix philosophy of distributed data. IEEE Data Eng. Bull., 38:4–14, 2015

    2015

  75. [76]

    The log: What every software engi- neer should know about real-time data’s unifying abstraction

    Jay Kreps. The log: What every software engi- neer should know about real-time data’s unifying abstraction. https://engineering.linkedin.com/ distributed-systems/log-what-every-software- engineer-should-know-about-real-time-datas- unifying, 2013

    2013

  76. [77]

    µFork: Supporting POSIX fork Within a Single-Address-Space OS

    John Alistair Kressel, Hugo Lefeuvre, and Pierre Olivier. µFork: Supporting POSIX fork Within a Single-Address-Space OS. In Proceedings of the ACM SIGOPS 31st Symposium on Operating Systems Prin- ciples, pages 18–35, 2025

    2025

  77. [78]

    Paxos Made Simple

    Leslie Lamport. Paxos Made Simple. ACM Sigact News, 32(4):18–25, 2001

    2001

  78. [79]

    Safe Agent Writes in Workbench Demo

    James Leng. Safe Agent Writes in Workbench Demo. https://www.dolthub.com/blog/2025-11-25- safe-agent-writes-video/ , 2025

    2025

  79. [80]

    AI-Assisted Engineering for Data Stream- ing with Lenses MCP

    Lenses.io. AI-Assisted Engineering for Data Stream- ing with Lenses MCP. https://lenses.io/kafka-ai

  80. [81]

    Lenses MCP: A New Era in AI En- ablement for Streaming Application Development

    lenses.io. Lenses MCP: A New Era in AI En- ablement for Streaming Application Development. https://lenses.io/blog/2025/10/lenses-mcp-new- era-in-ai-enablement-for-streaming-app-dev/

    2025

Showing first 80 references.