Derek Lau@xdereklau · 2026/08/16观点输出today i learned about runescape bench
having agents compete in complex games with vague paths and objectives (as opposed to a test with answers) is a very interesting way of benchmarking them
the ultimate test will be fully agentic businesses
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Derek Lau@xdereklau · 2026/08/05观点输出@ProofOfPlay thanks for giving it a good shot - all the best to the team 🫡
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Derek Lau@xdereklau · 2026/07/24观点输出@Skylinee yep crazy stuff
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Derek Lau@xdereklau · 2026/07/22政策影响so you’re telling me…
> openAI were doing a cyber security test
> the AI broke out of its sandbox and got access to the internet
> it then hacked another company (huggingface) to find the answers
> said company had to use chinese AI models to contain the exploit
??
AI：偏政策推进，强调合规落地与规则明晰，对监管主线更敏感。
Derek Lau@xdereklau · 2026/07/17观点输出@ape_rture will do and glad to see others share a similar view
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Derek Lau@xdereklau · 2026/07/17赛道影响"build a company that makes $100,000 profit. make no mistakes"
the output of this prompt is probably one of the best evals of models, since business is an infinite game
so i was thinking -- what does it take to build an agentic (or mostly agentic) company that is actually good?
my current mental framework for autonomous companies is that it's basically only as good as the stack you give it
in order to get good outcomes, you need to go deep into building specialised playbooks and harnesses for agents
e.g. which will get better outcomes: an agent that specialises in getting backlinks vs an agent that does SEO vs an agent that 'does marketing'?
not to mention different companies have different thresholds on what they would or wouldn't want to automate
implementing AI in a company isn't a tech problem, its a tech x domain expertise x ways of working problem
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
Derek Lau@xdereklau · 2026/07/06观点输出@Pleun_V i the opportunity is definitely there right now but there will also be other people offering the same service so you’ll need some point of differentiation or focus
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
Derek Lau@xdereklau · 2026/07/06赛道影响@gabusch @YGG_Play @LOLLandGame @waifusweeper best of luck with whats next and to the team affected
AI：偏主题投资表达，释放对基础设施与新叙事的偏多信号。
Derek Lau@xdereklau · 2026/07/06赛道影响one of the joys of not working is having time to zoom out and think about where all the AI hype is going (rather than just using it)
here’s what i have so far:
AI is not a product
it’s a tool to do things better, faster and cheaper
this used to be a trilemma (pick 2, never 3)
but the “trilemma” is shrinking due to a combination of foundational models getting better and cheaper, tailored rules & edge case handling for every vertical / niche under the sun, and general knowledge how to use the technology
this all leads to increased productivity for businesses, meaning software will get increasingly commoditised and cheaper for consumers
new types of products / experiences / workflows that were previously uneconomic or impossible now become possible
where value will go long term:
- the “winners” of foundational models
- SMB + enterprise SaaS startups that can build up a moat (data, tech entrenchment)
- infra that powers the AI / agents
- net new products / services
where value will go short term
- AI knowledge and execution (content, education, agencies, forward deployment)
- “inputs” to AI startups (data, knowledge, distribution)
- “normal” companies that find ways to meaningfully leverage AI before competitors
brand, human authenticity, taste, and distribution will rise in value as anyone can build anything
this is all assuming nothing blows up (eg. openai or anthropic can meet their > $1.5 trillion of compute and hardware commitments in the coming years)
Derek Lau@xdereklau · 2026/06/29观点输出@paytkaleiwahea @Dragoonneth someone pick this guy up
AI：延续个人公开立场，强调长期主义、执行效率与行业方向判断。
评论
暂无评论