The game’s title screen: Building 101 at DTU Lyngby at night, lights on, pine trees either side

Choose your own keynote What the audience did

Phil Ewels · scverse 2026 · DTU Lyngby · 12 October 2026

The keynote was a pixel-art game, and the audience steered it from their phones. 64 phones ranked the rooms, they sent around 630 messages, and asked 22 questions. 11 were answered on stage. The other 11 are at the bottom of this page.

The route

The talk opened in the lecture theatre. Phil then walked to three more rooms, in this order, with the live ranking showing on screen.

  1. The lecture theatre scene from the game
    1

    Why bother writing software?

    Lecture theatre

  2. The cloakroom scene from the game
    2

    Building tools for agents

    Cloakroom

  3. The coffee break scene from the game
    3

    Side projects

    Coffee break

  4. The courtyard scene from the game
    4

    rewrites.bio

    Courtyard

The foyer, with a doorway for each room

Doors left closed

Final ranking for the three rooms the talk never reached. Each phone gave its first choice the most points (Borda count).

  1. Community building with nf-core 133
  2. MultiQC 127
  3. Building awesome docs sites 124

Minute by minute

Messages from phones each minute: reactions, ranking changes, upvotes and questions. The busiest minute was 09:49, while people ranked the rooms before the start. During the talk, the busiest minute was 10:18.

Messages from phones per minute, with each question as it was asked and answeredRanking the rooms05010015009:4509:5009:5510:0010:0510:1010:1510:2010:2510:30Q&AQ&AQ&AAskedAnswered
Show the numbers
TimeMessagesQuestions
09:450
09:460
09:4712
09:4897
09:49134
09:5044
09:5112
09:529
09:5313
09:5413
09:557
09:569
09:5720
09:587
09:591
10:0091
10:017
10:0241
10:031
10:046
10:0551
10:066
10:074
10:0831
10:0981
10:1010
10:1181
10:1271
10:1310
10:142
10:1512
10:16172
10:1772
10:18246
10:19131
10:2012
10:21152
10:22101
10:2321
10:24121
10:254
10:262
10:274
10:282
10:291
10:306
  • Thumbs up
  • Love it
  • Celebrate
  • Laugh

These four reactions went straight to the stage and were never saved, so there is no count per emoji. They are inside the bars above. The message counts come from Cloudflare’s sampled analytics, so read them as close estimates.

Every question

All 22 questions, in the order they arrived, with the names people typed on their phones. Times are CEST.

11Answered on stage

  1. 10:00

    Does nfcore have some quantitative insights on agents actually using the llm.txt? Do they actually discover them over the docs?

    Tim

  2. 10:02

    How do we make agents build reusable blocks, rather than throwaway projects?

    Jesko1

  3. 10:05

    Does it make sense to write CLI tools for agents instead of just libraries? Why should an agent use CLI blocks rather than library blocks?

    Tomaz

  4. 10:08

    Do you feel that nf-core will get less popular as all LLM's default to python?

    Andris

  5. 10:09

    Because of agentic coding, a lot of labs are building their own tools. Do you think this is good? How can we standardize across labs?

    Jamie Bergen

  6. 10:11

    How do you manage your time to keep up with all the projects, and keep yourself updated with all the newest and greatest?

    Feifei

  7. 10:12

    How can we ensure that agentic tools are accessible to everyone, even people who don't have access LLM subscriptions?

    Michaela

  8. 10:16

    There are now agents that write the code and agents that review and plan the steps. Any advice on how to get humans in flow?

    Dhruva

  9. 10:16

    Where should human expertise lie when agents can thread blocks and frankly write better code? If supervision, why can't biologist take over?

    Agent_01

  10. 10:17

    How do you think about IP/scientific attribution? Original algorithm, new tool/maintainer, agent?

    Lucas

  11. 10:17

    Unlike nextflow, LLMs are inherently interactive and stepwise (with all its good and bad consequences). How can we close that gap?

    Jonas

11Not reached in time

  1. 10:18

    How much time does it take to prepare a presentation like this?

    Gaja1

  2. 10:18

    How sustainable do you think is agentic coding will be, once the current AI funding wave goeoess down?

    Ajay2

  3. 10:18

    How do you see hackathons evolve where nowadays people solve the project in 2 hours and then just eat cake for 3 days?

    Lukas1

  4. 10:18

    With rewrite, do you think at some point it will be only the agents who understand the code and not humans? Who'll be the maintainer?agent?

    Amirhossein

  5. 10:18

    some research institutions are worried about data privacy and “giving ideas away” to LLMs. do you have any thoughts on that?

    dani1

  6. 10:18

    How much code are you reading now? How are you verifying output etc

    Logan

  7. 10:19

    With the rapid evolution of AI and biological foundation models, how do you see bioinformatics evolving over the next five years ?

    Franz1

  8. 10:21

    Makes sense for already experts in the field, but what’s the value of getting a bioinformatics education in that reality? If any

    agent02

  9. 10:21

    How about an interpreted version of nextflow for easier learning with instant feedback and integrating script development with wf-developmen

    Dominik

  10. 10:22

    If code functionality is what matters (more than who writes code) should scientists be increasingly using CI/CD from day 1 while developing?

    Sid

  11. 10:24

    Do you see any problem junior bioinf/developers starting full on agentic coding from the get go wrt to their development?

    E

The game’s closing scene: an empty stage with a microphone stand under a spotlight

Thanks!

The keynote is still online. Walk the rooms the audience never reached, and look for the easter eggs.

Play the keynote