A Startup Says It Beat Microsoft and Google at Their Own Game

 Sourcetable, an AI-powered spreadsheet platform, put out a press release claiming its self-built "Sourcetable Benchmark" shows it scoring 24 out of 24 on a 24-test suite, ahead of Microsoft Copilot's 19 and Google Sheets' 17. The company says the widest gap showed up in tests requiring the AI to reach outside its own sandbox — connecting to a live database or pulling a file from the internet — where it claims Copilot and Google Sheets both scored zero. 

Key Points

• Sourcetable published a self-administered "Sourcetable Benchmark" claiming a 24/24 score against Microsoft Copilot (19/24) and Google Sheets (17/24)
• The results come from Sourcetable's own press release, not an independent third party, and should be read as a vendor claim rather than verified fact
• The starkest gap was in "connectors" — pulling live data from outside services — where Sourcetable claims 2-for-2 against zero for both incumbents
• This fits a broader, well-documented pattern: AI-native startups built without legacy architecture are increasingly out-executing incumbents on integration and execution speed
• For legacy software giants, the risk isn't losing a benchmark — it's losing the "who ships fastest" race in categories AI can now touch directly

Worth saying plainly: this is a company grading its own test and publishing the results as marketing. That doesn't make the underlying claims false, but it means the numbers deserve the same scrutiny we'd give any vendor-published benchmark until an independent party runs it. Sourcetable says the test suite and methodology are public and has invited Microsoft and Google to publish their own scores — a credible move if it happens, and one worth watching.

The Pattern Behind the Press Release

Whether or not Sourcetable's specific numbers hold up under outside review, the shape of the claim points at something real. Smaller, AI-native companies keep showing up with fewer legacy constraints, tighter build cycles, and a willingness to wire an AI model directly into live external systems — databases, APIs, financial data feeds — rather than layering a chat assistant on top of an existing product. Sourcetable's pitch is built entirely around this: connectors to Postgres, Salesforce, and Stripe, plus financial data pulled from sources like FRED and SEC filings.

Legacy platforms carry a different kind of weight. Microsoft and Google are shipping AI features into products used by hundreds of millions of people, which means every change moves through security review, backward compatibility, and enterprise trust commitments a five-person startup doesn't have to think about. That's not a knock on the products — it's a structural fact. It also means a startup with a narrower, purpose-built tool can genuinely outrun a giant on a specific, well-defined task, even if the giant would win on almost everything else the product needs to do at scale.

What This Could Mean for Software's Biggest Markets

If this pattern keeps repeating across spreadsheets, CRM, project management, and other mature software categories, the risk to incumbents isn't a single lost benchmark — it's death by a thousand narrow losses. A startup doesn't need to beat Excel at everything to peel off the users who only care about one thing Excel does badly. Enough of those losses, in enough categories, is how entrenched platforms lose ground they assumed was permanent.

That's the calculation teams should be running as they build out their AI adoption strategy: the competitive threat to legacy tools isn't necessarily a better version of the same product, it's a narrower tool that does one job with fewer constraints. For growth leaders evaluating where to place bets, that argument belongs in any growth strategy conversation about vendor selection — the newest, smallest player in a category is worth a real look, not just a courtesy demo.

None of this means Microsoft or Google are in trouble tomorrow. It means the moat that used to come from scale and distribution is thinner than it used to be, and vendor claims — even self-graded ones — are worth reading as an early signal, not proof.