Industries / Technology / Software and SaaS / Low-code platforms
Data and connections
Storing the records an app works on, and linking it to other software so data moves between systems without someone retyping it.
6
Milestones achieved, from 2011 to 2024.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2011
A weekend project links apps
Wade Foster, Bryan Helmig and Mike Knoop built the first version of Zapier in October 2011 at the first Startup Weekend in Columbia, Missouri, and won it. Zapier connects web apps through their public interfaces, so that an event in one, such as a new form entry, triggers an action in another. Zapier now says it connects AI tools to more than 9,000 apps.
Credit
Wade Foster, Bryan Helmig, Mike Knoop
2015
A spreadsheet with a database
Airtable raised $3 million in February 2015 for a tool that combines "the ease-of-use of spreadsheets with the power of relational databases", in TechCrunch's words. A relational database keeps records in linked tables, so one customer record can point to many orders.
Credit
Airtable and its founder Howie Liu
2016
Genes renamed for Excel
Excel, used with default settings, turns some gene names into dates, so SEPT2 becomes 2-Sep. Mark Ziemann, Yotam Eren and Assam El-Osta found the errors in 704 of 3,597 papers (19.6%) with Excel gene lists in 18 journals. In 2020 the HUGO Gene Nomenclature Committee renamed the affected genes, so SEPT1 became SEPTIN1 and MARCH1 became MARCHF1. A 2021 follow-up found errors in 30.9% of 11,117 papers. In October 2023 Microsoft let users switch off automatic conversions.
Lessona tool's default that silently changes data spreads until the tool lets people turn it off.
genomebiology.biomedcentral.compmc.ncbi.nlm.nih.govjournals.plos.orgtechcommunity.microsoft.com
Credit
Ziemann, Eren and El-Osta; HGNC; Microsoft's Excel team
2020
Cases lost to row limits
Public Health England said 15,841 COVID-19 cases from 25 September to 2 October 2020 were left out of its daily figures. The Register reported that lab results were stored in Excel's older.XLS format, which holds 65,536 rows per sheet; officials called it a "file size" issue.
Lessona file format chosen for convenience can cap a data pipeline without anyone noticing.
Credit
The Register, reporting PHE's statement
2020
Excel formulas become a language
In December 2020 Microsoft announced LAMBDA, which lets users define new functions in Excel's own formula language, including functions that call themselves. Microsoft Research wrote that "with LAMBDA, Excel has become Turing-complete", meaning any computation can in principle be written in formulas.
Credit
Andy Gordon and Simon Peyton Jones (Calc Intelligence project, Microsoft Research Cambridge) and the Excel team
2024
One plug for AI tools
On 25 November 2024 Anthropic released the Model Context Protocol (MCP), an open standard for connecting AI models to data and tools. In December 2025 the Linux Foundation took it into a new Agentic AI Foundation and said there were more than 10,000 published MCP servers, with the protocol adopted by Claude, ChatGPT, Gemini, Microsoft Copilot, Cursor and VS Code. Automation platforms now offer their connectors this way: Zapier MCP lets an AI client run actions in more than 9,000 apps.
Credit
Anthropic; the Linux Foundation
Watch for
An agent that edits spreadsheets
Microsoft says Agent Mode in Excel, which plans and makes changes in a workbook from a typed request, completed 57.2% of the 912 tasks in SpreadsheetBench, a public test of spreadsheet work, against 71.3% for people. Microsoft ran the test itself. A leaderboard kept by Shortcut, a rival spreadsheet agent, lists the 57.2% as "Unverified."
Achieved whenthe benchmark's authors or another party with no stake rerun Agent Mode on the full task set and publish the outputs.
Claimed by
Microsoft, September 2025
Open
01
What import does to data
A public test file of 200 values that software tends to change without asking: gene symbols such as SEPT2 and MARCH1, ZIP codes with leading zeros, ID numbers of 16 or more digits, dates in day-month and month-day order, fractions such as 1/2, strings such as 1E5 that look like scientific notation, and words such as TRUE. Load it by pasting and by uploading a CSV file into ten tools: Excel, Google Sheets, LibreOffice Calc, Apple Numbers, Airtable, Smartsheet, Notion databases, Microsoft Dataverse (the data store behind Power Apps), AppSheet and NocoDB. Record every value each tool changes. Rerun every six months, because the tools' behaviour drifts.
The briefCredit · unclaimed02
How often connectors break
A count of how often the integrations in automation tools break because another company changed its service. Use the public code histories of n8n, whose code anyone can read under its own "fair-code" license, and Activepieces, which is open source. Find each fix to an integration, sort the causes (the other service changed its interface, a bug in the integration, a new feature), and publish a breakage rate per integration per year and the median number of days from first report to fix.
The briefCredit · unclaimed
Open 01 · the brief
What import does to data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test file of 200 values that software tends to change without asking: gene symbols such as SEPT2 and MARCH1, ZIP codes with leading zeros, ID numbers of 16 or more digits, dates in day-month and month-day order, fractions such as 1/2, strings such as 1E5 that look like scientific notation, and words such as TRUE. Load it by pasting and by uploading a CSV file into ten tools: Excel, Google Sheets, LibreOffice Calc, Apple Numbers, Airtable, Smartsheet, Notion databases, Microsoft Dataverse (the data store behind Power Apps), AppSheet and NocoDB. Record every value each tool changes. Rerun every six months, because the tools' behaviour drifts.
Why it is still open
The closest test, by Abeysooriya and colleagues (2021), checked gene names only, in four spreadsheets: Excel and Google Sheets converted them, LibreOffice and Gnumeric did not. In 2016 Ziemann's team had found that Google Sheets converted none, so a single test goes out of date. Pollock (2023), a benchmark of 16 programs that load CSV files, states "we do not measure whether data types are inferred correctly."
Built with what exists
free tiers or trials of the ten tools; LibreOffice run without its screen, and Gnumeric's ssconvert, for scripted runs; Playwright to drive the web tools; Excel's Automatic Data Conversion settings, recorded on and off for each run.
First step
Write the first 50 values and load them into Excel, Google Sheets and LibreOffice both ways.
Done looks like
200 values, 10 tools and 2 ways of loading, 4,000 recorded outcomes, with the file and the scripts published. A second person reruns two tools from the scripts and matches at least 98% of the outcomes. Each maker gets its results 30 days before publication.
Open 02 · the brief
How often connectors break
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A count of how often the integrations in automation tools break because another company changed its service. Use the public code histories of n8n, whose code anyone can read under its own "fair-code" license, and Activepieces, which is open source. Find each fix to an integration, sort the causes (the other service changed its interface, a bug in the integration, a new feature), and publish a breakage rate per integration per year and the median number of days from first report to fix.
Why it is still open
A 2026 review of 97 studies of breaking changes covered Maven, npm, Python, web APIs and Linux distributions, and no workflow-tool connectors. The earlier study of how web API changes hit their users, by Tiago Espinha, Andy Zaidman and Hans-Gerhard Gross (2014), interviewed six developers about four APIs.
Built with what exists
the n8n repository (about 300 integrations, which it calls nodes) and the Activepieces repository (about 750, called pieces); PyDriller to read commit histories; the GitHub API or GH Archive for issues and pull requests; a small, cheap AI model to sort the fixes into causes, with a person checking a sample (see the Note "The jobs agents repeat").
First step
Label by hand every fix to ten popular n8n integrations, such as Slack, Google Sheets and HubSpot.
Done looks like
every integration fix from 2021 to 2026 classified, with 200 chosen at random checked by a second person who agrees on at least 180. Rates, fix times, data and code are published.