Arvid Kahl's predictions
Public predictions Arvid Kahl (@arvidkahl) made from 2023-11 to 2026-11, each checked against what happened.
Oracle
2 came true · 4 mostly true · 2 partly true · 1 pending
By topic
- AI 71% of 6
- Business 100% of 1
- Tech 75% of 1
Scored predictions
- Mostly true2025-01-03· AI· due 2025
Agentic AI would be the single biggest force changing how software businesses are built and run in 2025.
Agentic coding tools like Claude Code took off in 2025; Arvid himself said he built Podscan largely with it. Agents outside coding had less impact.
Original (thebootstrappedfounder.com)anthropic.comyespress.io
- Mostly true2025-01-03· Tech
Capable language models would keep shrinking, moving more AI processing from the cloud onto local devices.
Apple opened its on-device foundation model to developers in 2025, and small open-weight models improved, though most heavy use stayed in the cloud.
Original (thebootstrappedfounder.com)machinelearning.apple.com
- Partly true2025-01-03· AI· due 2026
By 2026, anyone would be able to build with AI what a capable founder could build in four hours in 2025.
Vibe-coding tools let non-programmers ship simple apps, but 'anyone' overstates it; security, deployment and quality gaps still trip up non-technical builders.
- Partly true2024-09-10· AI· due 2025-09-10
Within the next year, I bet we will have a code-generating AI / LLM that will take a full git repo, a prompt to optimize and refactor, and will iterate for a few hours over code variants and measure performance + unit...
Long-running code-optimizing agents arrived (AlphaEvolve, Codex), but with 20-30% speedups, not 95%.
- Came true2023-12-29· AI
AI API prices would keep dropping as competition between model providers intensified.
Per-token prices fell sharply through 2024 and 2025; OpenAI's GPT-4o mini launched at a small fraction of GPT-3.5 Turbo's price, and rivals cut prices in step.
- Mostly true2023-12-29· AI· due 2024
The burst of AI tooling seen in 2023 would be significantly surpassed in 2024, with no slowdown.
2024 brought multimodal frontier models, reasoning models and record AI investment, though 'significantly trumped' is subjective.
- Mostly true2023-12-29· AI
AI chatbots would evolve from conversational novelties into agents that take on and complete tasks.
Task-running agents such as OpenAI's Operator and coding agents shipped in 2025, though reliability for general tasks remained limited.
- Came true2023-12-29· Business
AI would take over first-level customer support.
By early 2024 Klarna said its AI assistant handled two-thirds of customer service chats, and AI front-line support became standard in SaaS help desks.
Pending
Not resolved yet, so not counted.
- Pending2025-06-06· Business· due 2027
AI would push SaaS acquisition multiples down over the next couple of years, as buyers build cheap in-house alternatives instead of acquiring.
The window runs to roughly mid-2027; no clear multi-year data yet on small-SaaS sale multiples.
Arvid Kahl's maker profile →Data updated 2026-11-05
Compare with other makers
92%
44%
27%
20%
0%
100%
Frequently asked questions
How accurate are Arvid Kahl's predictions?
Arvid Kahl got 75% of 8 resolved predictions right (2 came true, 0 were wrong), with 1 still pending.
How is the score calculated?
Came true = 1, mostly true = 0.75, partly true = 0.5, wrong = 0, averaged over resolved predictions. Pending predictions aren't scored until they resolve.