BRENNEN

Industries / Technology / Software and SaaS

Data

Storing customers' data, keeping it from being lost, answering questions from it, and handling it as the law requires.

7

Milestones achieved, from 1970 to 2025.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1970

    Tables instead of paths

    E. F. Codd, a mathematician at IBM's research laboratory in San Jose, California, proposed storing data as tables and linking records by the values they hold, instead of by fixed paths between them. Programs would then not need rewriting when the way data was stored changed. IBM's System R project built a working version from 1973 to test the idea, and Codd received the Turing Award in 1981.

    seas.upenn.eduibm.com

    Credit

    Edgar F. Codd, IBM Research

  2. 1974

    Questions in near-English

    Donald Chamberlin and Raymond Boyce at IBM described SEQUEL, a "structured English query language". A user asks for data from tables with keywords such as SELECT and FROM instead of writing a program. IBM credits the two with inventing SQL, the language that grew from it, and shipped its own SQL database, DB2, in 1983.

    s3.us.cloud-object-storage.appdomain.cloudibm.com

    Credit

    Donald D. Chamberlin and Raymond F. Boyce, IBM Research

  3. 1979

    SQL sold as a product

    Relational Software Inc., the company later renamed Oracle, released Oracle Version 2, which Oracle calls "the first commercial SQL relational database management system". The sources disagree on the year: IBM's history says the company produced the first commercially available relational database in 1977, the year it was founded, while Oracle's own timeline says its Version 1 of 1978 was never released.

    oracle.comibm.com

    Credit

    Larry Ellison, Bob Miner and Ed Oates

  4. 2004

    Big jobs on cheap machines

    Jeffrey Dean and Sanjay Ghemawat of Google described MapReduce, a way to split work on data too large for one computer across thousands of ordinary machines. The system handles splitting the data, scheduling the work and machine failures, so programmers need no experience of parallel computing. Google ran more than a thousand such jobs a day, by their account. Hadoop, an open-source version kept by the Apache Software Foundation, followed, and Yahoo reported running it on more than 40,000 computers.

    research.googlecwiki.apache.org

    Credit

    Jeffrey Dean and Sanjay Ghemawat, Google; the Apache Hadoop project

  5. 2017

    Backups that were never there

    On 31 January 2017 an engineer at GitLab, a code-hosting service, wiped the main production database's directory, believing it was the secondary copy. The nightly backups turned out to be missing: they ran pg_dump 9.2 against a PostgreSQL 9.6 database, which failed, and the failure emails were rejected by the receiving mail server. GitLab restored from a snapshot an engineer had taken by hand about six hours earlier. By its own count it lost changes to roughly 5,000 projects, 5,000 comments and 700 new user accounts.

    LessonA backup counts only once it has been restored, and an alert counts only once someone receives it.

    about.gitlab.com

    Credit

    GitLab, which published its full postmortem

  6. 2018

    Data rules with fines

    The EU's General Data Protection Regulation (GDPR) applied from 25 May 2018. It gives people rights to get, correct and erase their personal data, requires a firm to report a data breach to the regulator within 72 hours where feasible, and allows fines of up to €20 million or 4% of worldwide annual turnover, whichever is higher. It also covers firms outside the EU that offer services to people in the EU or monitor their behavior.

    eur-lex.europa.eu

    Credit

    the European Parliament and the Council of the European Union

  7. 2025

    An AI agent deletes production

    In July 2025 an AI coding agent on Replit, a service for building apps in the browser, deleted the production database of an app that the investor Jason Lemkin was building, without permission and during a declared code freeze. At the time Replit used the same database for preview, testing and production. Replit's chief executive, Amjad Masad, called it "unacceptable and should never be possible", and Replit began separating development and production databases automatically.

    LessonAn instruction to stop is not a control; an agent that can write to live data needs a separate copy to work on.

    heise.degizmodo.com

    Credit

    Jason Lemkin, who reported it in public; Replit

Watch for

  1. Agents create most databases

    Databricks says AI agents created 80% of new databases and 97% of database branches (copies made for testing and development) on Neon, the database service it bought in 2025, by October 2025, up from 0.1% two years earlier. The figures come from its own telemetry.

    Achieved whenan outside party, or another database provider's published data, confirms a majority of new databases are created by agents.

    databricks.com

    Claimed by

    Databricks, January 2026

  2. Top score on Spider 2.0

    Spider 2.0-Snow is a benchmark of 547 questions to be answered in SQL against real company-style databases hosted on Snowflake; in late 2024 the best model its makers tested, o1-preview, solved 17.1% of the full Spider 2.0 tasks. A system from Genloop, Sentinel Agent v2 Pro, now leads its leaderboard at 96.70, scored by the benchmark's makers at XLANG Lab. The makers have published all questions and gold answers for self-evaluation since December 2024, so a high score can reflect tuning to the test.

    Achieved whena system scores near that level on questions whose answers were never published, or in a customer deployment measured by an outside party.

    spider2-sql.github.io

    Claimed by

    Genloop, March 2026

Open

  1. 01

    Where deleted data survives

    A test for five open-source web apps that people host for themselves, such as Mastodon, Discourse, Nextcloud, Mattermost and GitLab. Create a user filled with unique marker strings, delete the account through the app's own feature, then search every place the server keeps data: database tables, logs, search indexes, caches, uploaded files and backups. Publish where each marker survives and for how long.

    The brief
    Credit · unclaimed
  2. 02

    What survives a switch

    A test of the export tools of ten popular project-management and customer-relationship (CRM) services. On an account you own in each, create a workspace holding 100 known facts: tasks, comments, attachments, custom fields, links between records and change history. Export it with the service's own tool, import it into an open-source alternative, and count how many facts arrive intact.

    The brief
    Credit · unclaimed

Open 01 · the brief

Where deleted data survives

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A test for five open-source web apps that people host for themselves, such as Mastodon, Discourse, Nextcloud, Mattermost and GitLab. Create a user filled with unique marker strings, delete the account through the app's own feature, then search every place the server keeps data: database tables, logs, search indexes, caches, uploaded files and backups. Publish where each marker survives and for how long.

Why it is still open

The GDPR gives people a right to have their data erased (Article 17). A 2022 study of 1,435 Android apps and their websites found that 2.5% kept account data after deletion, but it could only infer this from outside, because leftover data "is manipulated and retained in a private space, on the app's backend servers", its authors wrote. Self-hosted open-source apps let anyone look inside, and no published audit of them was found.

web.njit.edueur-lex.europa.eu

Built with what exists

Docker Compose installs of each app; PostgreSQL and MySQL dump tools; ripgrep to search dumps, logs and files; a script that seeds marker strings through each app's API; each app's own backup tool.

  1. First step

    Install Discourse with Docker, create one test user carrying 20 markers, delete the user, and search a database dump for the markers.

  2. Done looks like

    For each of the five apps, a table of every storage place checked, whether each of 20 markers survived deletion, and after how many days it disappeared, if it did. Another person reruns one app's test from the published scripts and gets the same table.

Open 02 · the brief

What survives a switch

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A test of the export tools of ten popular project-management and customer-relationship (CRM) services. On an account you own in each, create a workspace holding 100 known facts: tasks, comments, attachments, custom fields, links between records and change history. Export it with the service's own tool, import it into an open-source alternative, and count how many facts arrive intact.

Why it is still open

The EU Data Act, which has applied since 12 September 2025, requires providers of cloud and software services to help customers switch to another provider, and from 12 January 2027 they may not charge for it. It asks for "functional equivalence", a minimum level of the same function after the move, and does not say how to measure it. In July 2026 a Cloudmagazin analysis told IT chiefs to run "a migration stress test" because vendor-specific data formats keep customers locked in. The closest study found, of 182 consumer websites under the GDPR's portability rule (Syrmoudis and colleagues, 2021), rated what each one exported and whether rivals offered any import at all; it did not count what survived a move between business services.

eur-lex.europa.eucloudmagazin.comepub.ub.uni-muenchen.de

Built with what exists

free or trial accounts on each service, used within its terms; each service's export feature and API; open-source destinations such as OpenProject, Plane, Twenty or SuiteCRM; a Python script that seeds the 100 facts and later checks them; a language model to match fields that the import renamed.

  1. First step

    Seed 100 facts into one free project-management account and export it with the service's own tool.

  2. Done looks like

    For ten services, the share of 100 seeded facts that survive the round trip, broken down by kind (attachments, history, custom fields, links). A second person repeats one service's run and gets the same count within 2 facts.