Bilingual

My AI Adoption Journey
10 hours ago
- Adopting a new tool involves three phases: inefficiency, adequacy, and life-altering discovery.
- Chatbots are limited for coding; use agents that can read files, execute programs, and make HTTP requests.
- Reproduce manual work with agents to build expertise, breaking tasks into clear, actionable steps.
- End-of-day agents can perform research, triage issues, and explore ideas while you are unavailable.
- Outsource high-confidence tasks to agents while focusing on deep work, turning off notifications to avoid context switching.
- Engineer tools and prompts to prevent agent mistakes, improving reliability through 'harness engineering'.
- Aim to always have an agent running for delegated tasks, balancing it with manual work.
Understanding the AI Economy
5 days ago
- AI's economic transformation requires societal collaboration, not automatic outcomes.
- Google launched the AI & Economy ATLAS (v1.0) with 15 million de-identified interactions across 150+ countries, 140 languages, 800 occupations, and 4,000 tasks.
- Workplace AI adoption spans 68% of occupations but is shallow, used for only ~21% of tasks per job on average.
- Most work-related AI use is collaborative (ideation, strategy) rather than automated; less than 10% of interactions fully automate tasks.
- Manual workers (e.g., auto technicians) use AI for diagnostics and are twice as likely to use multimodal AI.
- Over 86% of AI interactions occur outside work, aiding household tasks and administrative frictions not captured in standard metrics.
- Global AI adoption correlates with GDP per capita but shows exceptions in middle-income countries; English accounts for only a third of conversations.
Lawyers risk being sued for failing to use AI
9 days ago
- Lawyers in the UK may face negligence claims for not using AI tools if a reasonable professional in their position would have used them, according to a UK Jurisdiction Taskforce statement.
- The statement highlights potential liability for incorrect AI use, such as failing to vet tools, explain AI to clients, secure data, or identify output errors, as well as for companies regarding AI chatbots representing them.
- It provides examples, including a solicitor failing to advise on AI-assisted document review, a radiologist not using AI for tumor detection, and an auditor skipping AI for large-scale anomaly detection.
- Critics argue that mandating AI use creates a double bind where lawyers are liable both for not using AI and for any undetected AI errors, undermining human judgment and risking inefficiency if tools produce inaccuracies.
- The taskforce, a government-backed body, aims to clarify how English law applies to new technologies, with previous statements on cryptoassets and smart contracts adopted by courts, while firms invest in AI tools like contract review platforms.
Australia Tops Claude Usage
15 days ago
- Claude AI is used six times more than expected in Australia, with homework assistance being the top use case at nearly 10%.
- Anthropic is exploring adding local infrastructure in Australia to meet data residency demands from enterprises and government agencies.
- Australia's 'values overlap' with Anthropic includes technological leadership and a strong, creative developer community.
- Anthropic signed a Memorandum of Understanding with Australia, committing to data centre expectations and strong copyright laws.
- Globally, Claude usage varies: personal support peaks weekends, business tasks dominate weekdays, and automation preferences differ by country.
Microsoft 365 Copilot adoption is under 4.5%, only 1% use it weekly
21 days ago
- Microsoft 365 Copilot has low adoption, with less than 5% of commercial customers paying for it, and only 20-30% of those users open it weekly.
- The paid Microsoft 365 Copilot add-on costs $21-$30 per user per month on top of existing subscriptions, offering priority AI access, work data integration, and agents like Researcher.
- Microsoft is adding third-party models like Claude to Copilot's paid tier, indicating low confidence in its own AI model, while relying on deep app integration as a selling point.
- Copilot faces reputation issues due to forced integrations into apps like Notepad and File Explorer, leading to user backlash and the nickname 'Microslop'.
- Microsoft is restructuring Copilot amid competition from Gemini and Claude, with web market share around 1% and GitHub Copilot losing mindshare to rivals like Cursor.
Companies hire more after AI adoption
22 days ago
- Companies adopting AI with high intensity grow headcount by 10% over two years, driven by firms in the top third of AI spending.
- Entry-level hiring increases by 12% among high-intensity AI adopters, suggesting a preference for candidates skilled in AI use.
- AI adoption is unevenly distributed, favoring larger, VC-backed, engineering-intensive, and fast-growing firms, with gains influenced by network effects and learning curves.
- AI adoption requires a minimum threshold (around $30 per employee monthly initially) and a 6-12 month learning period before impacting hiring.
- Small businesses that adopt AI intensely may experience outsized growth by reducing fixed costs, enabling new revenue streams and increased hiring.
- Practical advice: Young job seekers should prioritize AI-using firms; engineers should learn AI technologies; business owners should persist with AI despite initial challenges.
Tokenmaxxing is dead, long live tokenmaxxing
a month ago
- Tokenmaxxing, initially seen as wasteful, may have been a deliberate strategy by executives to overcome resistance to AI adoption within organizations.
- The concept of 'compounding correctness' suggests that increased token usage can lead to better outcomes, potentially reviving tokenmaxxing incentives.
- In cybersecurity, models like Anthropic's Mythos show that spending more tokens can enhance security by outspending attackers on exploit discovery.
- The rise of 'loops' in agentic AI allows for continuous, unsupervised operation, further driving token consumption.
- Open models like GLM 5.2 offer cost advantages over frontier models, making token-intensive processes more economically viable.
- The 'software factory' or 'dark factory' vision envisions fully automated coding pipelines, potentially justifying high daily token spends.
- Government regulation, such as controlled access to models like Mythos, is influencing AI deployment and creating opaque winner-picking dynamics.
- AI speed improvements, like OpenAI's Jalapeño chip, could shift workflows from async to more synchronous, real-time operations.

Hasty Briefsbeta

ai adoption

My AI Adoption Journey

Understanding the AI Economy

Lawyers risk being sued for failing to use AI

Australia Tops Claude Usage

Microsoft 365 Copilot adoption is under 4.5%, only 1% use it weekly

Companies hire more after AI adoption

Tokenmaxxing is dead, long live tokenmaxxing