We Tried the New ChatGPT Agent to Calculate Our AI Index — Here's What Happened

With the launch of OpenAI’s built-in ChatGPT Agent, we decided to test whether this new feature could finally handle a real workflow we’ve struggled with before: updating the TheMarketAI.com AI SC 15 Index.

In previous attempts using regular ChatGPT, the process was far from smooth — plagued by hallucinations, formatting limitations, and manual patchwork. We eventually had to download data from Yahoo Finance, convert it to a PDF, and feed it back into the model to even get close. It worked, but barely.

So when Sam Altman announced ChatGPT Agents — capable of combining tools like file reading and web browsing — we thought: let’s see if it can finally automate this properly.

Here was the prompt:

Please obtain the most recent market caps of the following companies: CRDO, IONQ, CVLT, RMBS, CWAN, IDCC, ITRI, QBTS, SOUN, AI, BBAI, CCSI, MVIS, EGAN, AISPPut the market caps in a table and multiply by their weight in the attached Excel.Aggregate the weighted market caps of all 15 stocks.Divide the aggregate by $74.6Bn and multiply by 100

The agent replied exactly as you’d hope:

“Understood. I will first download and review the attached Excel file to extract the weights… then research current market caps… calculate the weighted market caps… and provide the updated index value…”

It then opened a browser (very cool), and began retrieving market cap data one by one. Surprisingly, it didn’t use Yahoo Finance or Nasdaq directly — it relied on broader web search. Still, it was systematic and grounded. In under 6 minutes, it had everything lined up — no hallucinations, no confusion.

The final result? An index score of 106.49 — a 6.5% jump from our base of 100! We initially forgot to instruct it to reweigh the original index, but we figured it out quickly and it took another 50 seconds to correct.

TheMarketAI.com Take:

Last week, in our piece Did OpenAI GPT Agent Just Wipe Out a Whole Generation of AI Agent Startups?, we discussed the possibility that OpenAI’s new Agent could upend large parts of the emerging agent startup landscape. Now that we’ve tested it ourselves, we can better understand why the announcement drew such strong reactions.

The Agent completed a multi-step financial calculation using live web data, file handling, and structured reasoning — and it did so quickly, accurately, and without friction. That’s no small feat, especially for tasks that previously required multiple prompts, external tools, or human corrections.

We’re not saying this eliminates the need for specialized tools or startups building deeper agent-based solutions. But it does raise the bar. When the default tools improve this quickly, it becomes harder for smaller companies to differentiate — or even stay relevant. That’s the risk we explored in our earlier this month on whether AI is moving too fast for serious products to stabilize before the ground shifts again.

In that context, OpenAI’s Agent isn’t just a feature. It’s another step in a pattern: core capabilities that used to require paid integrations are getting absorbed into the base layer of AI platforms — faster than many expected.