The Gilbane Advisor

Curated for content, computing, data, information, and digital experience professionals

Gilbane Advisor 2-25-26 — agent reliability, cognitive surrender

February 25, 2026 /

This week we feature articles from Stephan Rabanser, Sayash Kapoor, & Arvind Narayanan, and Alberto Romero.

Additional reading comes from Ethan Mollick, JA Westenberg, Daniel Kocot, and Steve Hedden.

News comes from Graphwise, Dataiku, Krisp, and Cockroach Labs.

Our next issue will be published on 3/11/26

All previous issues are available at https://gilbane.com/gilbane-advisor-index

Opinion / Analysis

Towards a science of AI agent reliability

Quantifying the capability-reliability gap

“Surprisingly, even though the lack of reliability of AI agents is well known, right now the AI industry doesn’t have good tools for measuring reliability, or even a good definition of reliability.”

Stephan Rabanser, Sayash Kapoor, and Arvind Narayanan decided to take a stab at a “comprehensive measurement of reliability”. Together with some additional researchers have they’ve made a sensible and very useful start and published a draft, which they summarize in this post. The complete draft can also be downloaded. (9 min)

https://www.normaltech.ai/p/new-paper-towards-a-science-of-ai

A new Wharton study on AI warns of a growing problem: cognitive surrender

Casual users should pay special attention

Alberto Romero: “The study introduces the concept of “cognitive surrender,” our tendency to adopt AI outputs with “minimal scrutiny,” overriding “both intuition and deliberation.” I’ve read it, and the findings, although unsurprising, are still quite scary.” 

You can download the full paper, Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender, with the included link. (7 min)

https://www.thealgorithmicbridge.com/p/a-new-wharton-study-on-ai-warns-of

More Reading

All Gilbane Advisor issues

Content Technology News

Graphwise announced the immediate availability of GraphRAG

Beyond vector-only RAG, knowledge graphs provide context and common sense to AI. The AI-workflow engine turns “Python prototypes” into production-grade systems.
https://graphwise.ai/news/new-graphrag-solution-moves-beyond-vector-only-rag-knowledge-graphs-provide-context-and-common-sense-to-ai/

Dataiku launches 575 Lab, its open source initiative for responsible AI

The 575 Lab will focus on delivering deployable tools that strengthen explainability, privacy, and governance across modern AI and agentic systems.
https://www.dataiku.com/press-releases/dataiku-launches-575-lab/

Krisp launches real-time Voice Translation SDK

Enables CX (customer experience) platform developers to embed real-time multilingual voice-to-voice translation into live customer conversations.
https://krisp.ai/blog/real-time-voice-translation-sdk/

300-node clusters now supported in CockroachDB

CockroachDB v25.4 supports 300-node clusters and 1PB of data per cluster at 2.2M tpmC on transactional workloads to improve scalability for AI-era applications.
https://www.cockroachlabs.com/blog/300-node-clusters-supported-cockroachdb/

All content technology news

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, data, web, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

Krisp launches real-time Voice Translation SDK

February 18, 2026 /

Krisp announced the launch of its Voice Translation SDK, enabling CX platform developers to embed real-time multilingual voice-to-voice translation into live customer conversations. The technology has been live in production CX environments since 2025 as part of Krisp’s Call Center AI platform, operating in customer conversations globally before its SDK release.

Real-time voice translation must operate on continuous audio streams where latency, accuracy and conversational flow are tightly linked. Systems must recognize diverse accents, perform reliably in noisy environments and preserve natural turn-taking.

Krisp’s Voice Translation SDK is engineered to balance these competing constraints in live, two-way conversations. It supports any combination of over 60 languages and is optimized for synchronous interactions where clarity and conversational continuity are critical. This enables multilingual interactions within live conversations without requiring human interpreters.

The SDK is available for Windows, macOS and Web developers, allowing integration into both native and browser-based applications. To improve performance in real-world conditions, Krisp applies local Noise Cancellation before audio is processed in the cloud, isolating the primary speaker and improving recognition accuracy. The SDK also supports custom vocabulary and domain-specific dictionaries, enabling teams to enforce terminology and maintain consistency across professional environments.

https://krisp.ai/blog/real-time-voice-translation-sdk/

Dataiku launches 575 Lab, its new open source initiative for responsible AI

February 18, 2026 /

As AI moves from pilots to business-critical deployment, the issue is no longer access. It’s trust. Open source tools support that trust by keeping core components inspectable and standardizable, enabling stronger oversight across modern AI systems. Today, Dataiku announced the launch of the 575 Lab, Dataiku’s Open Source Office. The 575 Lab will release two new open-source toolkits designed to help enterprises make AI systems more transparent, governable, and fit for real-world use.

The 575 Lab will focus on delivering deployable tools that strengthen explainability, privacy, and governance across modern AI and agentic systems. The two initial open-source projects will be: 

  • Agent Explainability Tools that will help teams trace and understand decision-making across multi-step agent workflows, making agent decisions transparent for data scientists, compliance teams, and end users.
  • Privacy-Preserving Proxies that will enable safer use of closed-source models by protecting sensitive data end-to-end, and that teams will be able to run locally.

Both projects will be designed to support responsible enterprise AI, with a focus on reliability, security, transparency, and explainability.

The 575 Lab is now available to the community of AI specialists, data scientists, and developers responsible for creating, deploying, and scaling AI agents and applications.

https://www.dataiku.com/press-releases/dataiku-launches-575-lab/

Graphwise announced the immediate availability of GraphRAG

February 17, 2026 /

Graphwise announced the availability of Graphwise GraphRAG, a low-code AI-workflow engine designed to turn “Python prototypes” into production-grade systems instantly. It is based on a trusted semantic layer that reduces hallucinations and delivers precise and verifiable answers. GraphRAG unites LLMs, enterprise data, structured knowledge, and multiple search methods to deliver transparent, verifiable, enterprise-ready answers. Unlike standard RAG that “flattens” data into chunks leading to lost relationships and hallucinations, GraphRAG treats the knowledge graph as a trusted semantic backbone, ensuring AI responses are grounded in verifiable enterprise facts and complex relationships. Graphwise bridges the gap between complex enterprise data and functional AI agents. Features include:

  • Low-Code Visual Engine democratizes AI, enabling subject matter experts to adjust AI logic visually.
  • Out-of-the-Box Templates provide guardrails and support query expansion that deliver the fastest time-to-value.
  • Semantic Metadata Control Plane eliminates hallucinations and improves AI accuracy. AI responses are grounded in an organization’s “enterprise truth,” reducing risk.
  • Explainability and Provenance Panels support regulatory compliance. Built-in traceability affords transparency into how an AI response was produced.
  • Visual Debugging and Monitoring reduce maintenance costs by eliminating black box code.
  • SKOS-style Concept Enrichment harnesses domain-specific intelligence. This means AI understands company specific jargon, acronyms, and synonyms out-of-the-box.

https://graphwise.ai/news/new-graphrag-solution-moves-beyond-vector-only-rag-knowledge-graphs-provide-context-and-common-sense-to-ai

300-node clusters now supported in CockroachDB

February 12, 2026 /

From the CockroachDB Blog…

As AI-driven and agentic applications push data platforms into new territory, data architects are increasingly forced to choose between correctness, simplicity, and scale. To remove that tradeoff we’re announcing support for 300-node clusters with 2.2M tpmC and 1.2PB of data in CockroachDB v25.4.4 and beyond. Also, On CockroachDB Cloud, we’re announcing support for 64 vCPU per node. All customers will be able to self-serve and select these larger instance types if desired.

Highlights include:

  • ~610K QPS, which when compared to PUA on a 9-node cluster with 17K QPS shows that CockroachDB near linearly scales with the size of the cluster.
  • Compared to a previous run on 25.2, a run with the same amount of imported data on 25.4 took 30% less storage space than the previous run and enhanced compression.
  • Imports for this run on 25.4 were 2× faster compared to 25.1, for migrations to CockroachDB.
  • ADD COLUMN across 120 B rows completed without regression.
  • 330TB backup and 6 concurrent changefeeds completed in 2 hours and 40 min with no impact on foreground traffic.

Start with $400 in free credits. Or get a free 30-day trial of CockroachDB Enterprise on self-hosted environments.

https://www.cockroachlabs.com/blog/300-node-clusters-supported-cockroachdb

Gilbane Advisor 2-11-26 — multi-agent architectures, SaaS & CaaS

February 11, 2026 /

This week we feature articles from Nicole Königstein, and Scott Brinker.

Additional reading comes from JA Westenberg, Grace Huckins, Rahul Gaur, and Aruna Ranganathan & Xingqi Maggie Ye.

News comes from Elastic, Snowflake, Upland, and DeepL.

Our next issue will be published on 2/25/26

All previous issues are available at https://gilbane.com/gilbane-advisor-index

Opinion / Analysis

Designing effective multi-agent architectures

From models to systems

“… the winners in the agentic era won’t be those with the smartest instructions but the ones who build the most resilient collaboration structures. Agentic performance is an architectural outcome, not a prompting problem.”

Excellent advice from Nicole Königstein on how to get there. (8 min)

https://www.oreilly.com/radar/designing-effective-multi-agent-architectures

The SaaS moats are crumbling, but the opportunity is bigger

Context-as-a-Service (CaaS) and the future of domain-specific software platforms

An important read from Scott Brinker for all, not just marketers :). (8 min)

https://newsletter.chiefmartec.com/p/the-saas-moats-are-crumbling-but-the-opportunity-is-bigger

More Reading

All Gilbane Advisor issues

Content Technology News

Snowflake makes enterprise data AI-ready with Snowflake Postgres

The database now runs natively in the AI Data Cloud so enterprises can consolidate their transactional, analytical, and AI use cases onto a single, secure platform.
https://www.snowflake.com/en/news/press-releases/snowflake-makes-enterprise-data-ai-ready-with-snowflake-postgres-and-advanced-innovations-for-open-data-interoperability/

Elastic adds high-precision multilingual reranking to Elastic Inference Service

Two Jina reranker models and new Elastic Inference Service deliver low-latency, production-ready relevance for hybrid search and RAG workloads.
https://ir.elastic.co/news/news-details/2026/Elastic-Adds-High-Precision-Multilingual-Reranking-to-Elastic-Inference-Service-with-Jina-Models/default.aspx

Upland announces BA Insight Platform with integrated AI search experiences for enterprises

The new capabilities and enhancements are designed to make search experiences smarter, faster, and more insightful across complex enterprises.
https://investor.uplandsoftware.com/news/news-details/2026/New-Upland-BA-Insight-Platform-Delivers-Integrated-AI-Search-Experiences-for-Enterprises/default.aspx

DeepL launches voice API for real-time speech transcription and translation for instant multilingual communication

Developers can integrate real-time voice transcription and translation capabilities into their applications, significantly enhancing multilingual support for businesses.
https://www.deepl.com/en/press-release/deepl_launches_voice_api_for_real_time_speech_transcription_and_translation

All content technology news

The Gilbane Advisor is authored by Frank Gilbane and is ad-free, cost-free, and curated for content, computing, data, web, and digital experience technology and information professionals. We publish recommended articles and content technology news most Wednesdays. We do not sell or share personal data.

Subscribe | View online | Editorial policy | Privacy policy | Contact

Snowflake makes enterprise data AI-ready with Snowflake Postgres

February 3, 2026 /

Snowflake, an AI Data Cloud company, announced advancements that make data AI-ready by design, allowing enterprises to rely on data that is continuously available, usable, and governed as AI transitions from experimentation into production systems. With new enhancements to Snowflake Postgres, the database now runs natively in the AI Data Cloud so enterprises can consolidate their transactional, analytical, and AI use cases onto a single, secure platform. To help ensure AI systems are trusted at enterprise scale, Snowflake is embedding enhanced interoperability, governance, and resilience features into its platform.

Powered by pg_lake, a set of PostgreSQL extensions that allow Postgres to easily work within an organization’s open and interoperable lakehouse grounded in Apache Iceberg, enterprises can leverage Snowflake Postgres to directly query, manage, and write to Apache Iceberg tables using standard SQL. This capability is delivered within a Postgres environment, so enterprises can eliminate data movement between transactional and analytical systems.

Enterprises need data that remains open, governed, and resilient as it flows across engines, formats, and environments. Snowflake is expanding how customers access, share, and govern their data. Open Format Data Sharing extends Snowflake’s zero-ETL sharing model to include formats such as Apache Iceberg and Delta Lake.

https://www.snowflake.com/en/news/press-releases/snowflake-makes-enterprise-data-ai-ready-with-snowflake-postgres-and-advanced-innovations-for-open-data-interoperability

Elastic adds high-precision multilingual reranking to new Elastic Inference Service

February 3, 2026 /

Elastic, a Search AI Company, made two Jina Rerankers available on Elastic Inference Service (EIS), a GPU-accelerated inference-as-a-service that makes it easy to run fast, high-quality inference without complex setup or hosting. These rerankers bring low-latency, high-precision multilingual reranking to the Elastic ecosystem.

Rerankers improve search quality by reordering results based on semantic relevance, helping surface the most accurate matches for a query. They improve relevance across aggregated, multi-query results, without reindexing or pipeline changes. This makes them valuable for hybrid search, RAG, and context-engineering workflows where better context boosts downstream accuracy. The two new Jina reranker models are optimized for different production needs:

Jina Reranker v2 (jina-reranker-v2-base-multilingual)
Built for scalable, agentic workflows.

  • Low-latency inference with strong multilingual performance.
  • Ability to select relevant SQL tables and external functions that best match user queries..
  • Scores documents independently to handle arbitrarily large candidate sets.

Jina Reranker v3 (jina-reranker-v3)
Optimized for high-precision shortlist reranking.

  • Optimized for low-latency inference and efficient deployment in production settings.
  • Strong multilingual performance; maintains stable top-k rankings under permutation.
  • Cost-efficient, cross-document reranking: v3 reranks up to 64 documents together in a single inference call, reasoning across the full candidate set to improve ordering when results are similar or overlapping.

https://ir.elastic.co/news/news-details/2026/Elastic-Adds-High-Precision-Multilingual-Reranking-to-Elastic-Inference-Service-with-Jina-Models/default.aspx

« Older posts

Policies

Contact

twitter   rss   Mastadon   Bluesky   Medium

Contact • customerservice@gilbane.com

© 2026 Bluebill Advisors (except for contributed posts which are licensed under creative commons)

Banner photos: Sakonnet Harbor, Little Compton, RI, Fort Point Channel, Summer St bridge, and Northern Ave bridge, Boston

Bluebill Advisors Inc logo

© 2026 The Gilbane Advisor

Theme by Anders Noren — Up ↑