Five venture tasks
Representative founder work, such as comparing material claims and drafting partner outreach.
Paper 39 · Poster session
A Swappable, Local-First Agent Stack
Open-source agent stack and evaluation protocol for solo-founder social-impact ventures
Neil Carelse · Fri 23 Oct, 13:30–14:30 · Poster session, entrance area · LinkedIn (opens in new tab)
Research questions
- RQ1How should a swappable, local-first open-source agent stack for solo social-impact founders be decomposed into stable layers and replacement contracts?
- RQ2How can such a stack be evaluated comparatively across heterogeneous founder tasks without reducing success to benchmark accuracy alone?
- RQ3Under which task and deployment conditions does local-first infrastructure remain viable, and where are cloud escalation and human intervention still necessary?
Founderfinal authority
- 1Interfaces
- 2Orchestration
- 3Specialised agents
- 4Shared skills
- 5Inference routing
- 6Memory
- 7Security
- 8Infrastructure
- 9Product integration
Contribution
A nine-layer governed reference architecture
Each layer is replaceable behind a bounded contract, so a change is absorbed locally instead of forcing a full migration. The founder stays the permanent authority; agents are revocable delegates.
How the stack is evaluated
Three conditions
A deployment-comparison protocol, specified here and run in the associated thesis.
Beyond accuracy
Quality, strategic usefulness, reliability, oversight load, token efficiency, security behaviour, local completion and swap effort.
Presentation
Slides and talk
Slides prepared for this site; the paper is presented as a poster at AIS26.
Slide viewer
Loading…
Open full-screen talkDownload slides (PDF)
Poster
The poster
Loading…
Download poster (PDF, A0)
Paper
Read and cite
Carelse, N. (2026). A Swappable, Local-First Open-Source Agent Stack and Evaluation Protocol for Solo-Founder Social-Impact Ventures. In AI Summit Styria 2026 Conference Proceedings. FH JOANNEUM.
Accepted for the AI Summit Styria 2026 Conference Proceedings (FH JOANNEUM). Licensed under CC BY 4.0. DOI to follow.
Download PDF
Literature
Key sources
Nakara, W. A., Benmoussa, F.-Z. & Jaouen, A. (2021). Entrepreneurship and artificial intelligence: A systematic review and research agenda. Technological Forecasting and Social Change 166.
AI is becoming a capability multiplier in venture creation.
AI & entrepreneurship
Obschonka, M. et al. (2025). Artificial intelligence and entrepreneurship: A call for research. Entrepreneurship Theory and Practice 49(3).
Calls for research on how AI changes what founders can do.
AI & entrepreneurship
Chalmers, D., MacKenzie, N. G. & Carter, S. (2021). Artificial intelligence and entrepreneurship: Implications for venture creation in the fourth industrial revolution. Entrepreneurship Theory and Practice 45(5).
AI reshapes venture creation and the founder’s role.
AI & entrepreneurship
Ganuthula, V. R. R. (2025). AI-enabled individual entrepreneurship theory. Journal of Innovation and Entrepreneurship 14.
One founder can reach team-like scale with AI.
AI & entrepreneurship
Townsend, D. M. et al. (2025). Are the futures computable? Knightian uncertainty and artificial intelligence. Academy of Management Review 50(2).
Prediction cannot replace entrepreneurial judgement under deep uncertainty.
AI & entrepreneurship
Wang, L. et al. (2024). A survey on large language model based autonomous agents. Frontiers of Computer Science 18(6).
The four-module agent framework (profile, memory, planning, action) behind the layer split.
Agent architectures
Poll
Have your say
- Which of your work would you only let an AI process on a model you run yourself, not in the cloud?
- How much should an AI agent decide on its own in a small venture?
Vote at menti.com/alm8p7rokqe5
Get in touch