Databricks
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 19
- Sigma earned $200M, but says it never wanted to be BI
A commentary republished by 36Kr traces Sigma Computing, which sells analytics and BI tooling that runs queries directly on cloud data warehouses: it reports $200M in annual recurring revenue and more than 2,000 customers, yet both its founder and its current CEO have said for eight years that "the world doesn't need another BI tool," and the website now bills it as an "AI application platform." The piece recounts the product path from a 2018 spreadsheet-style interface to 2023's Input Tables write-back feature, followed by AI Q&A, custom agents and a code notebook, plus a May Series E of $80M with Databricks Ventures joining and Snowflake as an earlier investor. The author notes Sigma has never disclosed revenue by product line, that Gartner has placed it in the Niche Players quadrant of its BI Magic Quadrant for two years running, and that Sigma Tables, its bid to become part of the database layer, is still in beta — so the growth still comes from dashboards.
36氪 人工智能 · 🔥 9 - Databricks injects domain knowledge into prompts to improve LLM npm malware triage
A Databricks blog post describes findings and fixes from the LLM triage stage of its npm malware scanning pipeline. Collecting samples since November 2025, the team found that of 11,413 package-versions corroborated through OSV npm malware advisories, 96.2% hide their payload in the package source or fetch it from an external server, while only 3.8% (432 package-versions) bundle a binary. Those binary-bearing packages are the hardest to judge: among OSV-confirmed samples they failed to reach a decisive malicious verdict 14.4% of the time, more than double the 6.2% rate for source-only packages. The team built five adversarial Node.js native addon (.node) payload types covering credential theft, dropper/backdoor, reverse shell, cryptominer stub and persistence, with an obfuscation ladder running from a readable baseline to stripped symbols and single-byte XOR-encoded strings, in PE, ELF (all stripped) and Mach-O formats. Rather than fine-tuning models, they inject pre-computed static evidence into the prompt before inference, which they say improves detection quality for both budget and frontier models.
Databricks Blog · 🔥 8 - What Is an Agent Harness? The Scaffolding Around LLMs, Explained
According to a DeveDigest report, the term "Agent Harness" has become a buzzword in AI development after DeepSeek released an open-source desktop client called DeepSeek Harness and OpenAI launched an Agents API at its developer conference. The article explains that the model supplies only reasoning, while the harness is the surrounding system — execution loop, tool calling, context and memory management, sandbox permissions and replayable logs — that lets an agent actually finish real work. It cites testing by data analytics company Databricks finding that the same model plugged into different coding-agent harnesses can differ by more than 2x in per-task cost with no change in task quality, partly because a harness's system prompts and tool declarations consume tokens.
DoNews · 🔥 5 - Databricks details how its AI research team uses AI Runtime for agent-led training
Databricks published a post describing how its AI research team uses its own AI Runtime (AIR) to train and evaluate the models behind products such as Genie, Smart Routing and AI Functions. The team cites AIR capabilities including GPU preflight checks and health monitoring, cordoning off unhealthy nodes, the AIR CLI with custom Docker containers, automatic MLflow experiment tracking, governed data access through Unity Catalog FUSE mounting, and workload prioritization and preemption to share GPUs across projects. Databricks says agent-led "autoresearch" workflows have driven an 8x increase in the number of training jobs for its AI Functions research, with models ranging from a few billion to nearly a trillion parameters trained on NVIDIA H100 or B300 GPUs, and the largest runs using hundreds of GPUs over days or weeks.
Databricks Blog · 🔥 4 - TechCrunch Disrupt 2026 wraps with agent security and AI sessions
TechCrunch Disrupt 2026 runs October 13–15 at San Francisco's Moscone West, with October 15 as the final day; the organizer says 10,000+ founders, investors and technology leaders attend and 320+ startups exhibit. Day 3 includes several AI sessions, such as OpenAI's Alexander Embiricos on "What Comes After the Chatbot," a talk on agent security with Okta and NanoCo, Hugging Face's Irene Solaiman on how to know AI can actually do the job, and enterprise AI discussions involving Databricks and AMD. The day also features the Startup Battlefield final and track sessions on startups, fundraising and growth; passes are still being sold at 75% off the original price per the organizer (promotional).
TechCrunch AI · 🔥 3 - Databricks launches Predictive AI Functions in beta
Databricks announced that Predictive AI Functions are now in beta, offering SQL-native functions that let analysts and business users run forecasting, driver analysis and anomaly detection directly in SQL, dashboards and Genie without building or managing models. The launch adds four functions — ai_top_drivers, ai_detect_anomalies, ai_predict_value and ai_predict_class — and improves ai_forecast, which is powered by a pretrained transformer-based time-series foundation model and needs no tuning, with support for holidays and external covariates. Databricks says ai_forecast outperforms a leading cloud data warehouse's equivalent and its own previous version on the fev-bench benchmark, though the comparison vendor is not named; within Genie only ai_top_drivers is available today, with the others listed as coming soon.
Databricks Blog · 🔥 0 - FactSet cuts model training and deployment from days to hours with Databricks AI Runtime
A Databricks blog case study says FactSet's Text-to-Formula (TTF) system, which converts natural-language requests into FactSet Query Language formulas, initially relied on retrieval-augmented generation but hit a quality ceiling as queries grew more specialized, so the team added fine-tuned open models to its Compound AI architecture. FactSet says earlier fine-tuned models cut end-to-end latency by about 60%, and the team now fine-tunes models roughly every two to three weeks; using Databricks AI Runtime's serverless GPUs, it reduced the time to set up, train and deploy a custom TTF model from days to hours. The workflow runs alongside MLflow tracking and Lakehouse governance, with a reusable fine-tuning pipeline built on Lakeflow Jobs and Pipelines.
Databricks Blog · 🔥 0 - Databricks cuts Qwen 3.8 27B fine-tuning time 6x on AI Runtime
In a blog post, Databricks describes how systematic performance work on its AI Runtime cut fine-tuning time for Qwen 3.8 27B on a fixed dataset from 7 hours 44 minutes to 1 hour 17 minutes — about 6x — from a baseline throughput of 9,396 tok/s. The team notes that the initial run already showed 98% average GPU utilization, but that metric only measures whether the GPU is doing anything; the real bottleneck was hundreds of thousands of small ops that the CPU could not enqueue fast enough. Installing fused kernels such as flash-linear-attention and causal-conv1d brought training down to 2 hours 5 minutes (3.7x), and switching to 1D packed tensors with correct position_ids to cut padding waste delivered the final 6x gain, with example code published in an AI Runtime cookbook.
Databricks Blog · 🔥 0 - Databricks: AI agent ROI gap tied to missing business context, not speed
A Databricks blog interview with staff member Amber Roberts examines why enterprises are adopting AI agents quickly but struggling to prove ROI. She argues the buyers and the day-to-day users are often different people, and that buyers deploy under competitive pressure first, then spend six months to a few years learning which use cases actually work — the trial-and-error phase many organizations are in now. She says business context matters more than raw task speed, that the most common mistake is scaling before a single use case works well, and that traditional metrics like hours saved and tickets closed should be paired with AI-specific checks such as a third-party LLM judging correctness, hallucinations and tone. She also describes using two agents to contact and follow up with conference speakers, with all 137 speakers confirmed.
Databricks Blog · 🔥 0 - Flo Health fine-tunes open models with Databricks to boost medical safety and usefulness
A Databricks blog post describes how Flo Health, a women's health app with 81 million monthly active users, built safer and more useful custom open models on Databricks AI Runtime. Flo first created an evaluation system of 26 AI judges (12 medical-safety, 6 micro-behavior, 8 usefulness) to turn medical standards into a training signal, generated more than 15,000 privacy-first synthetic questions, then ran supervised fine-tuning and DPO on 32 NVIDIA H100 GPUs using LLM Foundry and Compose-RL with MLflow tracking. Flo reports medical safety rising from 34% to about 61%, usefulness from 42% to 76%, and the share of responses meeting both from 16% to 47%, and estimates tailored models can cut inference costs by an order of magnitude versus frontier APIs; it plans to apply the pipeline to newer open-weight models such as Gemma, Qwen and Kimi.
Databricks Blog · 🔥 0 - TinyFish runs its web-native agent model on Databricks AI Runtime
TinyFish is developing a specialized web-native model for enterprise agents that work across the live web, and has consolidated its post-training, evaluation and production serving on Databricks: managed GPUs via AI Runtime, evaluation with MLflow, and deployment through Model Serving. The team started with supervised fine-tuning and is moving toward Direct Preference Optimization. On its internal go/no-go benchmark, the model completed 1,691 of 2,178 web-agent tasks for a 77.6% task-success rate, with a separate LLM judge reviewing the request, action history and final result; partial completion or incorrect answers count as failures. Over a recent 15-day period, TinyFish recorded 294,781 agent runs involving the model across 6.4 million tool steps, at an estimated serving cost of $0.088 per run and $0.0040 per tool step — covering only that model, not the full agent stack. TinyFish already serves its search reranker and query rewriter on Databricks and uses Unity Gateway to centralize access to external foundation models.
Databricks Blog · 🔥 0 - Databricks Lakebase branches databases per coding agent
A Databricks blog post explains how Lakebase Postgres database branching supports coding-agent workflows: using copy-on-write, a full database branch can be created from a parent in under a second regardless of size, and branches scale to zero so idle ones incur no compute cost. The walkthrough gives each agent and each pull request an isolated ephemeral branch, combining Git worktrees, a post-checkout hook, GitHub Actions and Drizzle migrations, applying migrations to both a preview app and the main branch and posting a schema-diff comment on the PR. The author notes that Lakebase branches are not merged back to the main branch like Git; schema changes are promoted through migrations in code, and the speed and cost claims come from Databricks' own blog.
Databricks Blog · 🔥 0 - Anthropic launches Claude Dashboards and Claude Motion in beta
On October 8, Anthropic introduced two beta features for Claude. Claude Dashboards connects to data platforms such as BigQuery, Databricks, Snowflake, Amazon Redshift or ClickHouse, and to CRM tools like Salesforce, then builds a live dashboard from a plain-language question and keeps it current as the data changes; every number shows the query behind it, and results can be sent to Amplitude, Grafana, Hex, Mixpanel and other tools for deeper analysis. Dashboards is in beta on paid plans, while Claude Motion — which turns text, diagrams and images into short animated explainers built as editable code and exportable as MP4 — is in beta for Team and Enterprise plans. Anthropic also said Docs, Slides and Design have left beta and are available on every plan including Free, with more than 45 million documents created using those tools; the earlier Claude Design will be folded into the main app by December 14, with a migration guide published.
The Decoder · 🔥 0 - Europe's Q3 venture funding hits $25B, a four-year high, with AI taking 75%
European startups raised $25B in Q3, up 77% year over year and the region's strongest venture quarter in four years, accounting for 16% of global VC, according to Crunchbase data. AI companies took 75% of that, or $18.8B, a record share, with four rounds — Mistral's €3B Series D (about $3.5B), Nscale's $3.36B convertible note, Helsing's $1.8B and Quantum Systems' $1.2B — making up roughly 40% of the total. North America fell 35% quarter over quarter to $92B, largely because there were no new megarounds for OpenAI or Anthropic in the period.
Techmeme · 🔥 0 - US VC deal value hits record $515.8B as AI takes 82.7% share
US venture capital deal value reached $515.8 billion in the first nine months of the year, about 44% above the previous annual record, with AI accounting for a record 82.7% of that value, according to the PitchBook-NVCA Venture Monitor. Third-quarter deal value fell about 40% to $98.4 billion, and exits remain weak: SpaceX's $60 billion all-stock purchase of Cursor developer Anysphere alone made up 53.1% of quarterly exit value, leaving just $53 billion without it. PitchBook counts 179 new unicorns this year and 992 companies valued at $1 billion or more, and its model gives Anthropic an 86% chance of an IPO within a year, versus 12% for OpenAI.
SiliconANGLE AI · 🔥 0 - Omnigent open-source meta-harness orchestrates Claude Code, Codex and other agents
Omnigent is an open-source AI agent framework and "meta-harness" that, per its project documentation, mixes Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and user-defined YAML agents in the same session, with sessions that follow you between terminal, browser, phone and a macOS desktop app. The README also claims support for first-party API keys or Claude/ChatGPT subscriptions, cloud sandboxes such as Modal, Daytona, E2B and Kubernetes, and policies that pause for approval, cap spend or limit tool access, plus session sharing for teammates. The repository is currently listed on GitHub Trending for Python.
GitHub Trending · Python · 🔥 0 - Nvidia weighed last-minute OpenRouter bid before Stripe's $8B deal
According to The Information, Nvidia learned in July that AI model marketplace OpenRouter was being pursued for acquisition and told its leadership it was prepared to make a generous offer, but needed more time to evaluate a deal before making a formal bid. OpenRouter's founders did not want to wait, and Stripe ultimately prevailed with an $8 billion bid; Databricks and Stripe had already been in talks with the startup, and Nvidia's interest had not previously been reported. Over the following two months Nvidia signed more than $140 billion worth of deals, including a $105 billion credit guarantee, and held nearly $100 billion in equity investments plus $25 billion in future commitments as of late July.
The Information · 🔥 0 - Atlassian unveils Agentic Multiplayer Protocol, Rovo Work and Loom prompts
At Team '26 Europe, Atlassian announced the Agentic Multiplayer Protocol (AMP), which gives AI agents an admin-assigned identity, authority and scope so they can work alongside humans in shared tasks and context. It also introduced Rovo Work for long-horizon, multi-step tasks that humans review and approve, and pitched Loom as a way to turn narrated screen recordings into agent prompts. For developers, Rovo Code Search brings source code into the Teamwork Graph and Data Context extends to Databricks, Snowflake and BigQuery; the company says its expanded MCP server exposes 200 tools and handles roughly 15 million tool calls a day.
SiliconANGLE AI · 🔥 0 - TypeSafe AI’s Jev sparks interest in non-generative decision models
TypeSafe AI’s Jev is a decision model that maps inputs to predefined outputs such as yes/no answers, scores, or list selections instead of generating free-form text. CEO Diogo Almeida claims the model is used by about 25% of Fortune 500 companies and processed one trillion tokens per day roughly a week before the report, while separate reports say the startup is discussing a new funding round; these claims were not independently verified in the source.
IT之家 AI · 🔥 0
Experience and discussion from the community
Share my Databricks experienceAsk about Databricks
Nobody has shared their experience with Databricks yet.