Key findings
- Hex's Threads agent favors endorsed data and semantic models, and Explorers start in Endorsed Mode. Databricks tells Genie Agent curators to treat Genie as a new analyst who needs table descriptions, example SQL, and business rules.
- 28 of 61 public accounts from Hex users discuss its AI, 23 positively and 5 critically. Two of the critical reviews are about Magic, the assistant Hex replaced with the Notebook Agent in 2025.
- 12 accounts tie the quality of Hex's AI answers to the context it reads, whether a semantic layer, dbt metadata, written guides, or definitions a user keeps by hand.
- Chime launched each data domain only after Hex's agent passed that domain's set of test questions, its eval set. LangChain's data team had not adopted evals when it published, and relied on agent monitoring and its own review of important answers.
- Accounts split on who Hex serves. 6 call it a tool for data teams, and 7 describe business users across the company using it. 6 question whether it fits non-technical users.
Business users can rely on these answers once a data team curates what the agent reads and tests it against the questions people will ask. Hex and Databricks both document this method. Public accounts from Hex users show where rollouts keep that trust and where they lose it.
Both vendors put the data team in charge of context
Neither product answers well from the warehouse schema alone, so the data team's job moves from writing reports to writing definitions and keeping them current.
Hex's Threads agent heavily prioritizes semantic models and endorsed data, and by default Explorers can use Threads only in Endorsed Mode. Databricks tells Genie Agent curators to think of Genie as a new analyst joining the company, and to rely on table descriptions, SQL expressions, and example queries before text instructions.
Test before launch, then keep testing
Both vendors grade the agent against known questions before launch and keep grading it afterward. Hex runs eval suites written in YAML, scoring each answer against rubrics, from the CLI or on a schedule in Context Studio. Databricks runs benchmark question sets that score the accuracy of Genie's answers, and asks users to mark whether each answer is correct.
Chime made Hex's eval set the gate for every data domain. It reports that domain launch time fell from two and a half weeks to five days once the process ran on Hex's CLI.
Where users report trust breaking
The positive accounts describe curated setups. An analyst at a large company says Claude and Hex, connected to the warehouse and the semantic layer, find the right tables, write and run SQL, and check the results. A head of data says Hex's agents read the company's Snowflake warehouse, dbt metadata, and earlier Hex projects, and that the questions colleagues ask show where self-service falls short. LangChain's data team endorses only assets it has reviewed and asks staff to loop in the data team when an answer needs validation.
The critical accounts come from setups still in progress. A Reddit commenter who tried Hex and Omni side by side in January 2026 found Hex's agent harder to set up and had no good results from it yet. A Hex admin reports that non-technical colleagues take AI output as the source of truth even when it rests on wrong assumptions. The admin wants rules per user group that make them check it. A Product Hunt reviewer wants to see how calculations are produced before trusting them.
Methodology
GILD read Hex's and Databricks' documentation on October 12, 2026. User opinions come from 61 public accounts by Hex users, published from Jan 28, 2025 to Oct 2, 2026. Each count is the number of accounts that discuss a topic. The accounts are self-selected, so they show what users write about and do not measure Hex's customer base. Launch times and usage figures are the publishers' own, linked in Sources. Accuracy evidence comes from the cited teams' own use, Chime's eval gate included.
Products in this piece
Sources
- Official documentation
- 9
- Public source
- 9
- Hex docs: Threads
- Hex docs: Introduction to semantic modeling
- Hex docs: Evals
- Hex docs: Context Studio
- Hex docs: AI in Hex
- Hex customer story: Chime (Aug 4, 2026)
- Databricks docs: Tune Genie Agent quality
- Databricks docs: Test and monitor a Genie Agent
- Databricks docs: Create and manage a Genie Agent
- LangChain: How we built LangChain's agent-first data stack (Jul 27, 2026)
- Hacker News comment from a Hex workspace admin on context feedback (May 13, 2026)
- Reddit post from an analyst using Claude and Hex on a semantic layer (Aug 9, 2026)
- G2 review from a head of data on Hex's agents and Explorer seat price (Nov 24, 2025)
- Reddit comment on Hex Threads over the warehouse (Jan 20, 2026)
- Reddit comment comparing the Hex and Omni agents (Jan 21, 2026)
- Reddit comment from a fractional head of data comparing Hex, Omni, and Sigma (Jan 27, 2026)
- G2 review from a Hex admin on users treating AI output as the source of truth (May 8, 2026)
- Product Hunt review of Hex asking for transparent agent workflows (May 26, 2026)