Somewhere inside a company using OpenAI’s new enterprise tool, someone is about to type a single sentence into a chat window: “Why did sales drop last month?” Within seconds, an AI will pull real company data, build a dashboard, and hand back an answer — no analyst, no waiting for next week’s report, no learning a new piece of software.
That tool is called Data agent, and OpenAI has folded it into ChatGPT Work, the company’s business-tier subscription plan built for organizations rather than individual users. On the surface, it looks like just another AI feature. Underneath, it exposes a problem that has nothing to do with artificial intelligence and everything to do with human disagreement — and that problem is quietly becoming a paid service that almost no one is offering yet.
If you’re a freelancer, consultant, or solo operator anywhere in the world — whether you’re based in Bangkok, Ho Chi Minh City, Jakarta, or working remotely from a co-working space as a digital nomad — this is worth understanding closely. Not because you need to use Data agent yourself, but because the gap it reveals is a business opportunity.
What a “Data Agent” Actually Is
First, a distinction worth making. Most people think of AI chatbots as tools that answer short questions and then stop. An agent, in the way OpenAI and the wider AI industry now use the word, is different: it’s an AI system that takes a goal, breaks it into steps, and works through those steps on its own until the task is done — not just replying, but executing.
Data agent is built for a specific kind of task: answering business questions that normally require a human data analyst. Things like “why did sales drop,” “which expense category is growing fastest,” or “which major client is at risk of not renewing.” Traditionally, getting answers to these questions meant waiting for the next scheduled report or standing in line behind other requests to a company’s data team. With Data agent, an employee can open a new chat, type @Data followed by their question, and get an answer directly — no query-writing, no new analytics software to learn.
That sounds almost too convenient. And the catch is exactly where the opportunity lives.
The One Word That Breaks Everything: “Sales”
Here’s the problem OpenAI had to solve before Data agent could be trusted with real business questions: inside almost any company, the same word means different things to different departments.
Take “sales.” A sales team might count a deal as closed the moment a client signs a contract. The accounting department might not count it until the money actually lands in the bank. A product team might only count a customer as “sales” if they’re still actively using the service months later. Three departments, three different numbers, all called “sales.”
If an AI is left to guess which definition to use, it will inevitably get it right for one department and wrong for the others — instantly and with total confidence, which is worse than not answering at all. So OpenAI designed Data agent not to guess. Instead, it reads a company’s pre-agreed definitions: business terminology, how each metric is calculated, the formulas behind them, and how different data sets relate to one another.
Those definitions live in what’s called a semantic layer — a technical term for a layer of business logic that tells any system, human or AI, exactly which database table a term like “revenue” or “active customer” should be pulled from, and how it should be calculated. Data agent can read semantic layers built in systems like dbt (a popular tool data teams use to define and organize business logic), Databricks Genie Ontology, GitHub, and Snowflake Horizon, as well as dashboards already built in familiar reporting software like Power BI or Tableau.
The takeaway buried in this technical detail is the important one: Data agent’s usefulness has almost nothing to do with how smart the underlying AI model is. It’s capped entirely by how well a company has already agreed on its own definitions. OpenAI is candid about this — the internal teams that use Data agent successfully are the ones whose data teams had already built shared definitions and clear access rules before the AI arrived. There is no install-and-forget shortcut. If a company hasn’t sat down and agreed on what “sales” means, Data agent can’t fix that for them. Someone has to do that work first.
แนะนำ
No Special Access — The AI Only Sees What You’re Allowed to See
A natural fear with any “agent” is that it might quietly access information it shouldn’t. OpenAI’s answer here is reassuring in a way that also matters for anyone advising a business on AI adoption: Data agent has no special privileges of its own. Every action it takes runs strictly under the permissions of whichever employee account is connected to it.
Permission controls go down to a granular level — table, row, and even individual column. If an employee’s account can view a customer database but is blocked from seeing the phone number column, Data agent working on that employee’s behalf is blocked from seeing it too. So the real question isn’t “can the AI see something it shouldn’t” — it’s “what can this employee’s account see in the first place.” Every database connection also requires approval from a company’s system administrator; ordinary staff can’t wire up new data sources on their own.
Data agent connects directly to the databases companies already run — Amazon Redshift, Google BigQuery, ClickHouse, Databricks, MongoDB, Snowflake, and Datadog — plus documents stored in Google Drive or Microsoft’s SharePoint.
From Answer to Dashboard, and What Happens After
Once Data agent gives an answer, the interaction doesn’t have to stop there. Users can ask follow-up questions and immediately inspect the underlying evidence behind any conclusion — essentially clicking through to see exactly which data produced that number. This creates a real trade-off: take the summary at face value and move fast, or spend a few extra minutes tracing the source, so that when someone in a meeting asks “where did this number come from,” you can actually answer.
Data agent can then generate a full dashboard from its findings, complete with exploratory charts a team can edit, share, or refresh. If a team already opens Power BI every morning as part of their routine, Data agent can build the dashboard directly inside that tool instead of forcing everyone to switch to a chat window. Supported dashboard platforms include Omni, Oracle BI, Power BI, Sigma, Tableau, and ThoughtSpot.
It can even suggest next steps and who should be looped in, then send a summary via Slack or email, or trigger an action through a connected tool — but only actions a human has explicitly approved in advance. It doesn’t read the data and then decide to act on its own initiative.
The Freelance Opportunity Hiding in Plain Sight
Here’s where this stops being a story about enterprise software and starts being a story about income. OpenAI’s own admission — that Data agent is only as good as the definitions a company feeds it — describes a service gap that barely exists yet: helping small and mid-sized businesses define their own data before they can safely use any AI tool on it.
Large companies like Data agent’s early Alpha testers — NTT DATA (a major Japan-based IT services and consulting firm), Thermo Fisher (a US scientific and lab-equipment giant), and ServicePiston — already employ data teams to do this groundwork. Most small businesses, solo e-commerce sellers, and regional companies do not. That’s the market.
แนะนำ
For freelancers and consultants, this is a service you can package right now, long before Data agent even reaches general availability: AI-readiness auditing — sitting down with a small business owner and simply asking, “When you say ‘sales,’ what do you mean?” — then documenting the answer clearly enough that any future AI tool, chatbot, or dashboard can use it correctly. Add to that dashboard migration work (helping a business move from a messy spreadsheet habit to Power BI or Tableau) and basic KPI design for teams that have never formally agreed on what they’re measuring.
This is not a hypothetical trend. It mirrors what happened with earlier waves of business software: when spreadsheets, then CRMs, then automation tools like Zapier became mainstream, a wave of “spreadsheet consultants” and “CRM setup specialists” made steady income simply translating messy human processes into structured systems. Data definition and semantic-layer consulting is the next version of that same job, and almost nobody is marketing themselves for it yet.
How This Plays Out in Thailand and the Wider Region
For expats, digital nomads, and local entrepreneurs working across Southeast Asia, this opportunity has a distinctly regional shape. Thai small businesses that sell through Shopee and Lazada — the two dominant e-commerce marketplaces across the region, functioning much like Amazon — often track “sales” purely by marketplace payout dates, which rarely match what a business owner actually believes their revenue to be. Many small operators run customer service and even payments through LINE OA (LINE Official Account, a business tool built on LINE, the messaging app that dominates daily communication in Thailand, similar to how WhatsApp functions elsewhere), while collecting payments via PromptPay, Thailand’s instant bank-transfer system that lets customers pay a phone number or ID directly, without card fees.
None of these systems talk to each other automatically, and none of them agree on when a “sale” actually happened. That’s precisely the fragmented picture a semantic-layer consultant gets paid to untangle — and it’s work that can be found through Thai freelance marketplaces like Fastwork (a Thai equivalent of Upwork or Fiverr, widely used for hiring local freelance talent), then delivered remotely.
The same gap exists across Vietnam and Indonesia, where a similar boom in e-commerce, messaging-app commerce, and small-business digitization has left most companies with the same unresolved question: what do our numbers actually mean, and can we trust an AI to answer questions about them?
Borrow OpenAI’s Own Prompt Formula
Even without access to Data agent itself, there’s a technique buried in OpenAI’s launch materials that anyone can use today with any AI tool. OpenAI recommends three prompt types: finding the cause behind a number change, designing a KPI set for a new product, and summarizing monthly numbers for executives. What makes each one effective isn’t the topic — it’s that each prompt specifies exactly what the answer must contain, not just what’s being asked.
Asking “why did this number change” isn’t enough. A stronger version asks for the cause, a comparison to the previous period, and a recommendation on what to check next. Asking for an executive summary isn’t enough either — a stronger version demands the actual figures, the risks worth flagging, and a suggested next action. This single habit — stating the task and specifying the required components of the answer — instantly improves output from almost any AI tool, for anyone doing this kind of analysis work as a paid service.
What We Still Don’t Know
It’s worth being clear-eyed here: everything above comes from OpenAI’s own launch announcement. There’s no independent testing yet, no published pricing, and no confirmed timeline for general availability beyond the Alpha customers named. OpenAI says nearly every internal product team already uses Data agent, which signals real internal confidence — but that’s a different thing from proof it performs the same way once thousands of outside companies, with far messier data, start relying on it.
The Takeaway
Whether or not you ever touch Data agent yourself, the lesson underneath it applies immediately: AI tools are only as trustworthy as the human agreements sitting underneath them. For anyone building income around AI right now — consultant, freelancer, solo operator — the fastest-growing service isn’t prompting AI better. It’s sitting down with a business, however small, and getting everyone to agree on what their own numbers actually mean before any AI touches them. That work is unglamorous, it doesn’t require a coding background, and right now, almost nobody is offering it as a paid service. That won’t stay true for long.
Key Takeaways
- OpenAI’s new Data agent in ChatGPT Work answers business questions using a company’s own data, but its accuracy depends entirely on whether the company has agreed on clear definitions first.
- The AI has no special access of its own — it only sees what the connected employee’s account is already permitted to see, down to the column level.
- Answers come with visible sources, and results can become live dashboards inside tools like Power BI or Tableau.
- A real freelance opportunity exists in helping small businesses define their metrics and build semantic layers before they adopt AI tools — a service almost no one is marketing yet.
- This gap is especially visible in Thailand and Southeast Asia, where sales, messaging, and payments often run through separate, disconnected platforms.
Frequently Asked Questions
Q: What is OpenAI’s Data agent, and is it available to everyone yet?
A: Data agent is a business-data AI tool available inside ChatGPT Work, OpenAI’s enterprise plan. It’s currently in early rollout with select companies, and no public pricing or general release date has been announced.
Q: Can Data agent access any company data it wants?
A: No. It only works within the permissions of the employee account it’s connected to, and every new data connection must be approved by a company administrator first.
Q: Why does the same word like “sales” cause problems for AI tools?
A: Different departments often define shared terms differently — for example, sales teams may count a deal at signing while accounting counts it at payment — and an AI left to guess will confidently give an answer that’s wrong for at least one team.
Q: What is a semantic layer, and why does it matter for AI accuracy?
A: A semantic layer is a predefined set of business rules that tells any system exactly which data to pull and how to calculate it, and it’s the main thing determining whether an AI tool’s answers can be trusted.
Q: Does this mean small businesses can’t use AI data tools yet?
A: They can, but the tools work far better once a business has clearly documented its own metrics and terminology, which is often the missing step for smaller operations.
Q: Is there a business opportunity for freelancers in this?
A: Yes — helping small and mid-sized businesses define their metrics, clean up their dashboards, and prepare their data for AI tools is an emerging consulting niche with very little competition so far.
Q: How does this apply to businesses in Thailand specifically?
A: Thai businesses that sell through Shopee or Lazada and manage customers via LINE OA or PromptPay often have sales data spread across disconnected systems, making them strong candidates for this kind of data-definition consulting.
Q: What databases and tools can Data agent connect to?
A: It can connect to systems including Amazon Redshift, Google BigQuery, Snowflake, Databricks, MongoDB, ClickHouse, and Datadog, as well as documents in Google Drive or Microsoft SharePoint.
Q: Can Data agent take actions on its own, like sending emails or making changes?
A: It can send summaries or trigger connected actions, but only ones a human has already approved — it does not decide to act independently after analyzing data.
Q: Which companies are already testing Data agent?
A: OpenAI has named NTT DATA, Thermo Fisher, and ServicePiston as early Alpha testers, alongside near-universal internal use across OpenAI’s own product teams.
Q: What’s the best way to write prompts for this kind of AI data analysis?
A: Specify not just what you want to know, but exactly what elements the answer must include — such as the cause, a comparison to a previous period, and a recommended next step.
Q: Is this trend limited to Thailand, or does it apply elsewhere in Southeast Asia?
A: The same fragmented-data problem shows up across Vietnam, Indonesia, and other fast-digitizing markets, making this a regional opportunity rather than a Thailand-only one.
