I've seen plenty of companies announce AI pilots that look impressive in a presentation but never make it into everyday work. The harder question is whether an AI tool can become useful enough that employees rely on it for real tasks, with measurable results and proper oversight.
Oracle is offering a revealing example. In an October 8 customer story, OpenAI described how Oracle teams use ChatGPT Work and Codex for recruiting research, business reporting, software development, and incident response.
The headline numbers are striking: OpenAI reports 130,000 active ChatGPT users at Oracle, more than 95,000 active Codex users, and a 98% decrease in talent-acquisition research time. Those figures come from OpenAI's account of the deployment, but the underlying workflows provide useful lessons beyond the statistics.
Oracle's reported use cases combine AI-assisted research, business-data access, and engineering workflows.
How Oracle Says It Cut Recruiting Research Time
Oracle's talent-acquisition team built a market-intelligence tool using ChatGPT Work. It takes a job description, researches comparable roles, benchmarks compensation, and evaluates the talent pool across relevant locations.
That information previously took around two to four days to compile, according to OpenAI's case study. Oracle's recruiting team says preparation can now take about 15 to 20 minutes using the new workflow.
“We've built our talent market intelligence tool using ChatGPT Work, and it's really revolutionized how we do the front end of our recruitment process.”
Jan Ackerman, Senior Vice President and Global Head of Talent Acquisition at Oracle. Source: OpenAI's Oracle customer story.
The less obvious benefit is consistency. Instead of each recruiter approaching market research differently, a shared workflow gives hiring managers a more standardized starting point for conversations about role requirements and candidate availability.
That does not remove the need for judgment. Recruiters still need to validate compensation data, evaluate candidate suitability, and consider local hiring conditions that an automated summary might miss.
Codex Turns Business Questions Into Reports and Applications
Oracle Applications Lab works on systems and processes used across the company. The team built an internal map of business objects, their relationships, and the rules governing them—then used Codex to help translate plain-language questions into SQL queries and operational outputs.
A business user can describe the information they need rather than manually searching for a report or asking a specialist to assemble one. The workflow can retrieve information from internal systems and return an analysis, report, or application.
What the internal business-data model does
- Defines business context: Maps important objects, relationships, and rules.
- Connects questions to data: Uses that context to help translate requests into SQL and system calls.
- Produces usable outputs: Returns analysis, reports, or applications instead of only a conversational answer.
- Preserves validation: Requires the underlying data definitions and access rules to be correct.
This is arguably the most important lesson in the case study: the AI is not working from a vague prompt alone. Oracle invested in encoding the business context that makes its data meaningful in the first place.
Without that foundation, a language model might generate syntactically valid SQL that uses the wrong table, misunderstands a business term, or applies the wrong filter. A well-defined data model reduces ambiguity, although it does not eliminate the need to test query results.
Codex Also Helps Oracle Respond to Software Incidents
Oracle's production-engineering teams use Codex to gather relevant incident context and retrieve the appropriate response playbook. Instead of spending as much time hunting through documentation, site reliability engineers can focus more quickly on what happened and what to investigate next.
Oracle Applications Lab executive Richard Lam says a simple incident that previously took about an hour could be handled in minutes. That is an account of Oracle's workflow, not a guarantee that every incident or outage can be resolved that quickly.
Complex incidents still demand people who can interpret evidence, assess risk, coordinate responses, and decide when to roll back a change. An AI system that retrieves the wrong playbook or misses a critical dependency can make a difficult situation worse.
ChatGPT Work vs. Codex: Why Oracle Uses Both
| Area | ChatGPT Work | Codex |
|---|---|---|
| Primary focus | Research, analysis, documents, and business workflows | Software engineering and technical tasks |
| Oracle example | Talent-market research and recruiter preparation | Internal SQL workflows, applications, and incident context |
| Useful outputs | Research summaries, reports, and working documents | Code, queries, tests, and technical workflow outputs |
| Human responsibility | Check research quality and make decisions | Review code, architecture, security, and system behavior |
The distinction is not that one tool is intelligent and the other is not. They serve different kinds of work, and Oracle's examples show why businesses may need both a general work environment and a more technical coding agent.
ChatGPT Work
Designed for research, information analysis, and producing business artifacts such as documents, reports, and spreadsheets. Oracle's recruiting example shows how a repeatable workflow can reduce the time needed to prepare for hiring discussions.
Explore ChatGPT WorkCodex
Designed for software-development tasks such as writing and debugging code, running tests, and working with repositories. Oracle also uses it to support data-driven business workflows and gather context for engineering incidents.
Explore CodexThe Overlooked Lesson: AI Does Not Replace System Design
Oracle's own leadership emphasizes that employees remain responsible for the systems AI helps build. Lam cautions that teams need to pay close attention to architecture, security, and the way code is structured.
This is easy to miss when a case study emphasizes speed. A tool can produce a working prototype quickly, but an enterprise still needs maintainable code, controlled permissions, repeatable tests, monitoring, and a clear owner for each workflow.
Four practical steps before automating a workflow
- Measure the baseline: Record how long the task takes today, including review and rework.
- Provide real context: Connect approved documents, business definitions, and system rules instead of relying on a prompt alone.
- Restrict permissions: Give the agent only the data access and actions required for that task.
- Test the output: Compare results with known-good examples and keep a human reviewer for consequential decisions.
Oracle's examples are especially useful because they focus on concrete workflows rather than simply asking employees to use AI more often. The goal is a repeatable process with a defined input, clear permissions, a useful output, and a way to check whether it worked.
What the Adoption Numbers Do—and Don't—Tell Us
OpenAI's case study reports 130,000 active ChatGPT users and more than 95,000 active Codex users at Oracle. These figures suggest broad internal adoption, but the page does not provide a detailed methodology for the term “active” or an independent audit of usage.
The case study also doesn't establish that every Oracle employee uses the tools, that all deployments deliver similar savings, or that the company has eliminated the need for specialists. Its strongest evidence is the description of specific processes and how the teams say those processes changed.
What looks promising
- Time savings tied to defined workflows.
- More consistent recruiting preparation.
- Faster access to internal business information.
- Less time spent searching for incident context.
What still needs scrutiny
- Independent validation of savings figures.
- Accuracy across more task types and edge cases.
- Security, access controls, and data governance.
- Long-term maintenance and total operating cost.
Watch OpenAI's ChatGPT Work Walkthrough
OpenAI's product walkthrough shows how ChatGPT Work can coordinate multi-step tasks across connected tools and turn an objective into a reviewable output. It provides useful context for understanding the work-oriented approach used in Oracle's example.
Source video: OpenAI — Meet ChatGPT Work. This is a product walkthrough, not an independent audit of Oracle's results.
Final Takeaway: Start With the Workflow, Not the AI
Oracle's story suggests that enterprise AI can become more useful when companies connect the technology to a real process, supply the relevant business context, and measure the result. ChatGPT Work supports research and business tasks, while Codex helps with technical workflows and software engineering.
The reported 98% reduction in recruiting research time is striking, but the broader lesson is more durable: define the task, build the right context, apply appropriate guardrails, and evaluate the output. Speed matters, but it is only valuable when the result is accurate, secure, and maintainable.
Is AI Doing Too Much of Your Thinking?
Oracle just proved that enterprise AI can automate 98% of your research, but what happens to your brain when you stop doing the heavy lifting? Read our deep dive into ChatGPT and cognitive offloading to discover how outsourcing your daily workflows might be quietly impacting your critical thinking skills.
Read the Cognitive Impact Guide →Sources and further reading
OpenAI — How Oracle turns days of work into minutes with ChatGPT and Codex
Oracle — OpenAI selects Oracle Cloud Infrastructure to extend its AI platform
OpenAI — Access OpenAI models and Codex through Oracle Cloud
Frequently Asked Questions About Oracle and OpenAI
How does Oracle use ChatGPT Work?
Oracle's talent-acquisition team uses ChatGPT Work in a talent-market intelligence tool that researches comparable roles, benchmarks compensation, and assesses talent availability across locations to help prepare for hiring discussions.
Did ChatGPT reduce Oracle's recruiting research time by 98%?
OpenAI reports a 98% decrease in talent-acquisition research time. The case study says research that previously took two to four days can now be prepared in about 15 to 20 minutes, but the results are company-reported and not independently audited in the article.
How many Oracle employees use ChatGPT and Codex?
OpenAI's October 8, 2026, customer story reports 130,000 active ChatGPT users at Oracle and more than 95,000 active Codex users. The article does not provide a detailed independent audit of those usage figures.
How does Oracle use Codex for business analytics?
Oracle Applications Lab built a model of business objects, relationships, and rules to help Codex translate plain-language questions into SQL queries and return analyses, reports, or applications from internal systems.
Does Oracle use Codex without human oversight?
No. Oracle's account emphasizes that people remain responsible for system design, architecture, security, and the quality of the resulting code. AI-assisted workflows still require validation and ongoing maintenance.
No comments:
Post a Comment