Gov/en/Portal:R&D/Innovations:AI Supported Deployment: Difference between revisions
Visible in-content table of contents (Theo 2026-08-15) |
Terminology: libre/free software instead of open source (Theo directive 2026-09-20) |
||
| (2 intermediate revisions by the same user not shown) | |||
| Line 3: | Line 3: | ||
{{ExpertIntro|Methodology page, proposed as an initial hypothesis. Boris Cherny (Anthropic) describes five work archetypes emerging as AI blurs product roles: Prototyper, Builder, Sweeper, Grower, Maintainer. WikiDeal's R&D intends to adopt this framework, to be tested during Prototype 1: open calls would invite people matching these profiles to move from idea to prototype to deployment.}} | {{ExpertIntro|Methodology page, proposed as an initial hypothesis. Boris Cherny (Anthropic) describes five work archetypes emerging as AI blurs product roles: Prototyper, Builder, Sweeper, Grower, Maintainer. WikiDeal's R&D intends to adopt this framework, to be tested during Prototype 1: open calls would invite people matching these profiles to move from idea to prototype to deployment.}} | ||
{{ArticleSummary| | |||
* [[#context|Context: when AI blurs job titles]] | * [[#context|Context: when AI blurs job titles]] | ||
* [[#the-five-profiles|The five profiles]] | * [[#the-five-profiles|The five profiles]] | ||
| Line 11: | Line 11: | ||
* [[#state-of-the-art|State of the art]] | * [[#state-of-the-art|State of the art]] | ||
* [[#see-also|See also]] | * [[#see-also|See also]] | ||
}} | |||
__NOTOC__ | __NOTOC__ | ||
| Line 29: | Line 30: | ||
=== What recent studies say about AI acceleration === | === What recent studies say about AI acceleration === | ||
Recent empirical studies give a nuanced picture of how much AI actually accelerates a project. A controlled experiment with GitHub Copilot found that developers assisted by an AI pair programmer completed a coding task 55.8 percent faster than a control group (Peng et al., 2023). By contrast, a 2025 randomized controlled trial by the research organization METR, with 16 experienced | Recent empirical studies give a nuanced picture of how much AI actually accelerates a project. A controlled experiment with GitHub Copilot found that developers assisted by an AI pair programmer completed a coding task 55.8 percent faster than a control group (Peng et al., 2023). By contrast, a 2025 randomized controlled trial by the research organization METR, with 16 experienced free software developers working on mature projects they knew well, found that allowing early-2025 AI tools ''increased'' completion time by 19 percent, against the developers' own expectations. Anthropic's 2026 analysis of about 400,000 Claude Code sessions observed that people make most of the planning decisions (what to do) while the AI makes most of the execution decisions (how to do it), and that on coding tasks people from non-engineering occupations succeed at nearly the same rate as software engineers. | ||
Read together, these studies suggest that AI support can strongly boost a project but does not do so automatically: how the human roles are organized around the AI appears to matter. This is one of the reasons why WikiDeal intends to test an explicit organization of roles, described below, rather than assuming that AI alone would accelerate the work. | Read together, these studies suggest that AI support can strongly boost a project but does not do so automatically: how the human roles are organized around the AI appears to matter. This is one of the reasons why WikiDeal intends to test an explicit organization of roles, described below, rather than assuming that AI alone would accelerate the work. | ||
| Line 111: | Line 112: | ||
💡 '''Improve this concept:''' submit a proposal via [[Gov/en/Portal:R&D/Open-Calls:Main|Open Call]] | 💡 '''Improve this concept:''' submit a proposal via [[Gov/en/Portal:R&D/Open-Calls:Main|Open Call]] | ||
[[Category:Innovation]] | [[Category:Innovation]] | ||
<!-- visible --> | <!-- visible --> | ||