Data Agent on DataAgentX

Data Agent Weekly: Implementation Challenges at DACon26

Sun, 25 Oct 2026 00:00:00 +0000

Covering October 18–24, 2026 (Beijing Time). This issue focuses on six engineering challenges distilled from 19 sessions at the Beijing leg of DACon26, with other developments organized by product and open-source categories.

On October 23–24, DataFun hosted the Beijing leg of DACon26. The six issues below are synthesized from relevant official agenda items; the commentary in this article represents editorial technical analysis. Session titles used here are abbreviated; for full official titles and speakers, refer to the DataFun official agenda link in the Sources section at the end.

The 2W1H of the Semantic Layer for Agents: Why It's Needed, What It Is, and How to Build It

Fri, 23 Oct 2026 00:00:00 +0000

Does connecting to a database and generating SQL guarantee that an Agent will answer business questions correctly?

When asked the same question, “What was last quarter’s revenue?”, different tools yield different answers: a BI tool might include taxes and shipping costs; a Notebook might fail to deduct discounts and returns; finance may recognize revenue based on shipment dates, while the Agent is still guessing how tables relate.

Before Adjusting a Fee Rate, Identify Which Transactions Will Be Affected

Thu, 22 Oct 2026 00:00:00 +0000

If a payment fee rate is reduced from 14 basis points to 1 basis point, how much can a merchant save in a month?

While this appears to be a multiplication problem, the actual challenge lies before the multiplication: determining which transactions are affected. This analysis recalculates two “hard” problems from DABstep, with a focus on defining the scope and performing independent verification.

Hotel Data in Practice: Why Brand Rankings Change

Tue, 20 Oct 2026 00:00:00 +0000

Why doesn’t a brand with an average 5-star rating always rank first on the hotel brand leaderboard?

This walkthrough follows the complete pipeline using public benchmark questions: Data → Question → Solution → Validation. The focus is not merely on calculating the answer, but on understanding how qualification thresholds and denominator selection alter conclusions.

Data and Questions

The dataset is InfiAgent-DABench, which evaluates the ability to truly execute analysis and deliver results starting from CSV files, natural language questions, and constraints. This fixed question set contains 257 questions referencing 52 CSV files; here, we use only hotel_data.csv, comprising 1,057 rows and 9 columns, covering 7 cities and 103 non-null brands, with each row representing a hotel record.

Tencent Cloud DataBuddy: The First Signs of RSI in Data Agents

Tue, 20 Oct 2026 00:00:00 +0000

The Tencent Cloud DataBuddy team announced the new version of DataBuddy during a live stream on the DataFunTalk WeChat video account. The materials do not specify the broadcast date; this article was compiled on October 20, 2026.

Two points are particularly noteworthy: DataBuddy can connect to existing external data sources, not just those on Tencent Cloud; and in the future, models can also integrate proprietary APIs, such as those for Claude and ChatGPT.

Data Agent Weekly: What Are Teams Globally Building?

Sun, 18 Oct 2026 00:00:00 +0000

Covering October 11–17, 2026 (Beijing Time). This week begins with four domestic developments before moving to global progress.

Synthesizing these updates, ten signals point in the same direction: The competition for Data Agents is shifting from “can it answer” to “can it participate in work continuously and reliably.” This manifests as reusable applications, inheritable context, semantics preceding execution, and verification built into workflows.

Domestic: Turning Analysis Paths into Reusable Assets

01 Alibaba Cloud Quick BI: Attribution Logic and Asset Reuse (10/13)

Turning 15 Skills into an Agent You Can Verify

Sat, 17 Oct 2026 00:00:00 +0000

Does installing 15 Skills give you a reliable Data Agent?

One essential step remains: turning “what it can do” into “how to prove it has done the job well.”

This article introduces no new official Skill. It presents an approach to product development synthesized from the source material discussed in the previous 9 installments. The task protocols, acceptance chain, and failure checklist below are editorial frameworks, rather than a unified specification defined by the source.

What Is Published Is a Snapshot, Not a Real-Time System

Fri, 16 Oct 2026 00:00:00 +0000

Does publishing analysis results as a URL mean the team now has a real-time data system?

No. The original documentation for Publish Artifact to Sites clarifies this boundary first and foremost: what is published is a snapshot; it does not automatically add live data source connections.

This boundary dictates all subsequent design decisions: how access permissions are configured, how versions are managed, and what must be communicated during handoff.

Dashboards and Reports Share Evidence, but Serve Different Jobs

Thu, 15 Oct 2026 00:00:00 +0000

When a dashboard starts resembling a long article, or a report becomes a wall of metrics, the underlying problem often precedes visual design. Does the reader need to keep monitoring a situation, or understand a particular judgment?

Data Analytics Skills addresses these needs with two Skills: Build Dashboard and Build Report. They share a foundation of trustworthy evidence, but their purposes, structures, and acceptance criteria differ.

Reproducible Calculations Make Charts Credible

Wed, 14 Oct 2026 00:00:00 +0000

The same chart changes when someone else reruns the analysis. The same data tells a different story when the axis changes. Details like these often undermine a Data Agent’s credibility.

Data Analytics Skills divides this work between two Skills: Jupyter Notebooks makes the calculations reproducible, while Visualize Data keeps their presentation from misleading readers. The first explains where a number comes from; the second establishes what a chart can support.

Data Quality and Validation: Two Gates, Two Risks

Tue, 13 Oct 2026 00:00:00 +0000

If a table has no nulls and no duplicates, can you confidently publish the analysis?

Not necessarily. Even if the data is entirely correct, the calculation might be wrong; even if calculated correctly, the comparison might be flawed, or correlation might be mistaken for causation. These three types of failures occur at different stages and require distinct checks to catch them.

Data Analytics Skills divides these into two skills: Analyze Data Quality to check the data, and Validate Data to check the analysis. These two gates guard against different risks; missing either one leads to leakage.

How Big Is the Market? Deliver an Estimate You Can Interrogate

Mon, 12 Oct 2026 00:00:00 +0000

“What is the size of this market?” The most common response from a Data Agent is a number that appears precise. The hardest deliverable is an estimation framework that can be interrogated and adjusted.

The focus of the Market Sizing skill is not to pad answers with zeros, but to ensure scale estimations have boundaries, sources, and assumptions, while explicitly stating where evidence is insufficient when it is lacking.

When Metrics Drop: Reproduce the Change Before Discussing Causes and Actions

Sat, 10 Oct 2026 00:00:00 +0000

“Conversion rates dropped; the new version must be unusable.”

This statement skips two critical questions: Are the metrics truly comparable? Does the change genuinely stem from product experience? An investigation must first rule out changes in definitions, coverage, or denominators before attributing the movement to the business itself.

Data Analytics Skills divides this work into two Skills: Metric Diagnostics explains the change, while Product & Business Analysis translates that explanation into evidence-based business decisions. The former owns “what happened,” and the latter owns “what to do.”

From Defining Success to Sustained Operations: The Complete KPI System Loop

Fri, 09 Oct 2026 00:00:00 +0000

Many teams maintain a long list of KPIs yet cannot clarify three things: what success means, whether they are currently on track, and what the next step should be. As the number of metrics increases, judgment often deteriorates.

Data Analytics Skills breaks this down into two Skills: Design KPIs defines success, while KPI Reporting organizes actual performance, targets, drivers, and actions into operational readouts. Together, they form a closed loop from measurement contracts to sustained operations.

Gather Enough Business Context to Continue

Thu, 08 Oct 2026 00:00:00 +0000

“Help me understand why conversion fell.” If a data agent immediately finds tables, writes SQL, and slices channels, it may answer the wrong question very efficiently. What are the numerator and denominator? Does conversion mean registration, activation, or payment? Which window changed? What decision will the result support?

Gather Business Context puts the request back into business coordinates before computation begins. Its output is not a final diagnosis. It is a Context Note that makes the next analytical step safe to start.

Who Routes the 15 Data Analytics Skills?

Wed, 07 Oct 2026 00:00:00 +0000

A data agent may know how to design metrics, diagnose anomalies, inspect quality, run notebooks, draw charts, build dashboards, and publish reports. That breadth creates a subtle failure mode: every tool contributes, yet no component owns the actual question.

The Index in the Data Analytics Skills collection is interesting because it acts less like a homepage and more like a control plane. It decides whether a request belongs in analytics, which workflow owns it, how conflicting sources are handled, and what “done” means.

How to Build a Data Agent: A Working Map from 23 Articles

Sun, 04 Oct 2026 00:00:00 +0000

After surveying many Data Agent products, a team beginning implementation still faces five basic questions: What work should the system own? Which data should it use? How may it execute? How will we prove the result? Who remains accountable when it fails?

Those questions are more durable than a vendor list. They connect definition, architecture, autonomy, evaluation, and rollout.

The diagram below maps those questions onto an operating path: how a task enters the system, how semantics constrain planning and execution, and how evidence, approval, read-back, and feedback close the loop.

Why Volcengine Routes Data Q&A and Deep Research Separately

Fri, 02 Oct 2026 00:00:00 +0000

“What was East China revenue yesterday?” and “Should we change product, audience, or outreach after repurchase fell?” are both data questions, but they are different jobs. The first is bounded and quickly verifiable. The second requires decomposition, multiple evidence types, and conditional judgment.

Volcengine separates intelligent data Q&A from deep research. The point is not the number of agents; it is giving different responsibilities different execution paths.

Inside DataBuddy: Why Data Agents Need a Shared Data Foundation

Tue, 29 Sep 2026 00:00:00 +0000

An analytics agent reports 120,000 new users. The operating dashboard says 108,000. Both can show their SQL, and both claim to use current data. The problem is larger than one wrong answer: do the agent and the existing data products share the same assets, semantics, identity, and controls?

That question makes DataBuddy’s architecture more interesting than its three visible roles—data engineering, analytics, and governance agents. Underneath them, the product connects ingestion, lakehouse storage, catalog, unified semantics, and AI-ready data knowledge into a shared chain. An Agent Harness ties that chain to context, tools, security, and evaluation.

How a Data Agent Understands Average Order Value

Mon, 28 Sep 2026 00:00:00 +0000

“What is average order value by category over the last 30 days?” A model can easily produce valid SQL and an invalid business answer. Average order value may be paid amount divided by paid orders, exclude refunds, use buyers as the denominator, or count only fulfilled orders.

This is not primarily a SQL problem. The phrase has not yet been bound to executable meaning.

Why Microsoft Fabric Data Agents Stay Read-Only

Sat, 26 Sep 2026 00:00:00 +0000

Must an agent modify data or send tickets to be valuable? Fabric Data Agent takes a narrower position: it generates and executes read-only SQL, DAX, and KQL, leaving cross-system writes to external orchestration.

Identity comes before querying

Data access runs under the user’s Microsoft Entra ID and source permissions. Seeing a schema does not grant permission to read its data. Power BI semantic models remain subject to Read permission, RLS, and CLS. Enforcement stays in the data platform rather than in prompt instructions.

Three AWS Data Agent Routes Solve Different Problems

Thu, 24 Sep 2026 00:00:00 +0000

Amazon Quick, SageMaker Data Agent, and Amazon Bedrock are not three products to rank by feature count. They begin with different users and responsibility boundaries.

Route	Primary user	Main deliverable	Team responsibility
Amazon Quick	Business and BI users	Conversational answers, insights, stories, workflows	Topic semantics, access, answer review
SageMaker Data Agent	Analysts, data scientists, ML teams	Queries, code, analytical development	Environment, cost, code review, reproducibility
Amazon Bedrock	Application developers	Programmable agent capabilities	Tool contracts, identity, state, evaluation, UX

Quick: from Q&A toward business work

The current Quick experience emphasizes cross-dataset questions and agentic workflows, with multi-dataset Topics acting as a semantic layer. It is closest to a business application, but querying several datasets does not mean cross-domain metrics are already aligned. Topic design still controls terms, fields, and aggregation.

Google Cloud Data Agents Converge Three Kinds of Context

Tue, 22 Sep 2026 00:00:00 +0000

Knowing where a table lives is not the same as knowing how to answer a business question. Google Cloud describes Conversational Analytics data agents as a combination of context and tools: business information and data provide context; SQL, Python, and visualization tools perform work.

Three contexts have different authority

Data context identifies queryable objects, fields, relationships, and locations. Business context explains terms, metrics, filters, and verified queries. Conversation context retains the user’s scope, assumptions, and follow-up questions across turns.

Anthropic Skills Turn Analytics Practice into Workflows

Mon, 21 Sep 2026 00:00:00 +0000

Connecting a model to a warehouse does not create a dependable analyst. The model still needs the team’s methods for checking data, choosing calculations, presenting evidence, and reviewing a deliverable.

Anthropic defines a Skill as a folder of instructions, scripts, and resources loaded when relevant. That turns repeated prompt explanation into a versioned work protocol and allows deterministic code to handle steps where token generation is a poor substitute.

How Kuaishou Scaled Its Data Agent Beyond 10,000 Weekly Users

Sun, 20 Sep 2026 00:00:00 +0000

A data agent reaching more than 10,000 weekly active users matters because it has moved into data production, business diagnosis, and recurring delivery—not because a chat box received more prompts. Kuaishou’s September 19, 2026 technical presentation described a progression from bounded workflows to a unified agent entry point.

The presentation reported more than 10,000 weekly users, over 80 percent penetration in core data-analysis and data-operations roles, more than 2,000 Skills, and over 600,000 evaluation records. These are company-reported snapshots, not independently audited results or promises that transfer to another organization. The system relationships behind the numbers are more useful than the headline itself.

Inside OpenAI’s Data Agent: Context Before SQL

Fri, 18 Sep 2026 00:00:00 +0000

Two tables both contain user_id; why do they produce different active-user counts? Enterprise analytics often fails before SQL generation: the system chose the wrong table, timestamp, filter, or definition.

OpenAI’s January 2026 account of its internal data agent centers the design on context engineering. The article’s snapshot described more than 3,500 internal users, 600 PB of data, and roughly 70,000 datasets. The more transferable lesson is how the system narrows that search space.

A Data Agent Vendor Map Based on System Position, Not Labels

Thu, 17 Sep 2026 00:00:00 +0000

The easiest Data Agent market map to draw incorrectly is a wall of logos that mixes cloud platforms, BI applications, internal systems, open-source frameworks, and research projects. They may serve different jobs, occupy different layers, and accept different production responsibilities.

A useful map has two axes: system position and delivery model. Find peers first, then compare them.

The First Step in Enterprise Data Agent Adoption Is Not Model Selection

Tue, 15 Sep 2026 00:00:00 +0000

The first enterprise Data Agent decision is usually not the most capable model or the number of agents. Choose a frequent, verifiable, low-impact segment of work and define what correct completion looks like. Otherwise a smooth demo reaches production with no credible test of trust.

Rollout is a staged transfer of responsibility. Each step toward production gives the agent another duty and requires new evidence plus a return path.

How to Read Data Agent Benchmarks Before Trusting the Ranking

Mon, 14 Sep 2026 00:00:00 +0000

A first-place Data Agent is not automatically the best procurement choice. A leaderboard says which submissions performed better under a particular task set, data environment, tool policy, budget, and scoring rule. An enterprise needs to know whether the system can deliver under its definitions, permissions, cost, and failure conditions.

The most useful way to read a ranking is backward through its measurement chain.

OpenAI Data Analytics Skills as Reviewable Work Protocols

Sun, 13 Sep 2026 00:00:00 +0000

“Why did conversion fall?” A demo agent can quickly locate a table, write SQL, draw a line, and suggest several causes. Delivery exposes the harder questions: What is the denominator? Is the current period complete? Did channel mix change? Are the “causes” measured contributions or a plausible story?

The Data Analytics Skills are useful because they turn analytical responsibility, handoffs, and completion criteria into reusable work protocols—not because of one special prompt.

When Is a Data Agent Report Actually Delivered?

Sun, 13 Sep 2026 00:00:00 +0000

An agent has generated a report on declining new-customer conversion. The task is not finished. A business leader needs to know whether the conclusion supports a decision, an analyst needs to recompute material values, and an operations team may need an investigation ticket. They require different artifacts and may have different data rights.

Delivery is complete when an authorized recipient can understand, verify, and use the result. Recommendation, approval, submission, confirmed effect, and business impact must remain separate states.

How Far Data Agent Governance Must Extend

Sat, 12 Sep 2026 00:00:00 +0000

A regional manager and a corporate analyst can ask the same question—“How is channel conversion performing?”—and be entitled to different answers. Row-level filtering in the database is not enough if a cache, export, or shared report can later deliver the corporate result to the regional user.

Governance cannot be a system prompt that says “do not reveal sensitive information.” It must cover discovery, model context, tool invocation, runtime, delivery, and business actions, enforced by systems the model cannot override.

OpenAI Data Agent Connects Enterprise Data, Dashboards, and Action

Fri, 11 Sep 2026 00:00:00 +0000

On September 10, 2026, OpenAI launched Data Agent in ChatGPT Work. The release is more than a Text-to-SQL chat interface: it connects enterprise sources, business context, analytical workflows, interactive dashboards, and approved actions in one conversational path.

The competitive question shifts from whether a model can write a query to whether more employees can complete reviewable data work under existing organizational controls.

If the SQL Runs, Can You Trust the Data Agent Answer?

Thu, 10 Sep 2026 00:00:00 +0000

A successful query, attractive chart, and complete report do not independently make an answer trustworthy. The agent may have selected the wrong source, bound the wrong metric, executed correct SQL over incomplete data, and then described correlation as causation.

One accuracy score compresses these failures into a number and hides high-cost errors. A better evaluation follows the analysis chain layer by layer and sends each failure back to the component that can repair it.

Why Agent-Generated SQL Must Not Run Directly

Wed, 09 Sep 2026 00:00:00 +0000

A model that emits SQL has proposed an action; it has not earned production execution authority. Even a SELECT can scan terabytes, expose sensitive columns, invoke an external function, or return a plausible result after an incorrect join.

The execution layer converts model intent into a controlled operation: validate before running, constrain during execution, verify afterward, and retain evidence for every side effect.

Why Data Agent Tasks Still Fail When the Agent Can Plan

Wed, 09 Sep 2026 00:00:00 +0000

“New-customer conversion is down. Find the cause and recommend what to do.” The agent queries channels, then regions. The third query times out, so it restarts from the beginning. Or it produces a twelve-step plan, reads the first result, and forgets which hypothesis the task was meant to test. It appears to reason, yet the work never finishes.

The planning layer is not a checklist generator. It turns a complex objective into an execution process that preserves state, checks evidence, recovers from failure, and hands control to a person when necessary.

The Data Agent Access Layer: Finding the Right Data Before Querying

Tue, 08 Sep 2026 00:00:00 +0000

“Why did new-customer conversion fall this month?” A warehouse may contain a production funnel model, a similarly named test table, and a spreadsheet from last quarter’s campaign. All three are reachable. All three have plausible columns. If the agent chooses the wrong one, polished SQL and charts only make the error more persuasive.

Before query generation, the access layer must answer three questions: What may this identity access? What should this task use? Is the selected evidence sufficient for the claim? It is not a pipe that dumps every schema into a prompt. It is a permission-aware discovery and context-routing system.

Why Stronger Data Agents Need a Semantic Layer

Tue, 08 Sep 2026 00:00:00 +0000

Give a model the schemas for orders, payments, and customers and it can quickly produce executable SQL. Ask “What was revenue last month?” and the real work begins. Does revenue mean booked value, cash received, net receipts, or recognized revenue? Which date applies? In which period do refunds belong?

Those are not universal facts for a model to infer. They are enterprise contracts. As an agent gains tools and a longer execution chain, the cost of propagating one incorrect definition into a report or action increases.

A Seven-Layer Architecture for Production Data Agents

Mon, 07 Sep 2026 00:00:00 +0000

“Why did new-customer conversion fall this month?” sounds like a SQL question. It is not.

Before touching a warehouse, a useful Data Agent has to establish what new customer means, which conversion window applies, whether the metric changed recently, which data the requester is allowed to see, and what evidence would make an explanation credible. A query can run perfectly and still answer the wrong question.

Production Data Agents therefore need more than a language model and a database connector. They need a chain of responsibilities that turns ambiguous intent into a result another person can inspect. The seven-layer model below is a way to design that chain—and, when it breaks, to locate the failure.

Data Agent Levels L0–L5: Autonomy Transfers Responsibility

Sun, 06 Sep 2026 00:00:00 +0000

One system writes SQL. Another invokes SQL and Python, observes errors, and replans. The second system has more tools, but tool count alone does not make it more autonomous. The real questions are: Who sets the goal? Who organizes the work? Who is authorized to act? Who verifies the result? Who recovers—and remains accountable—when it fails?

Data Agent levels are therefore best understood as a transfer of responsibility, not a feature checklist. At each level, the agent inherits more operations, judgment, and risk.

Five Data Agent Routes: Compare Products by the Work They Own

Sat, 05 Sep 2026 00:00:00 +0000

One Data Agent turns a business question into SQL. Another runs queries and Python, then writes a report. A third edits a production pipeline. They share a category name, but they should not share one scorecard.

The useful unit of classification is a workflow, not a vendor. Ask four questions: What enters the system? Which part of the work does it own? What does it deliver? What is the blast radius when it fails? Those questions reveal five recurring routes.

The SQL Ran. Why Is the Answer Still Wrong?

Fri, 04 Sep 2026 00:00:00 +0000

The most dangerous result in analytics is often not a failed query. It is a query that runs cleanly, returns a precise number, produces a polished chart—and answers the wrong question.

Suppose an operator asks, “How many new customers did we acquire this month?” The system finds a customer table, counts distinct IDs by created_at, and returns 12,486. The query is reproducible and syntactically correct. But Sales defines a new customer as a first-time payer, while Finance excludes refunded transactions and internal test accounts. The query counted new customer records, not the metric that will drive the budget decision.

What Is a Data Agent? Four Practical Tests

Thu, 03 Sep 2026 00:00:00 +0000

“Why did conversion fall this week?”

An NL2SQL tool can translate that sentence into a query. A data agent has a larger job. It must establish which product and funnel the question refers to, decide whether “this week” means a calendar week or a rolling seven-day window, and respect what the current user is allowed to see. It then has to find trusted data, apply the right metric definition, investigate the change, run and check its queries, and return evidence with the conclusion.