Anthropic has weird vibes

Introduction and Personal Experience 00:00

  • Anthropic is praised by developers for their models that simplify code writing and for their Cloud Code product.
  • Despite technical strengths, Anthropic is criticized for not being good stewards of open source and software development.
  • The presenter has spent $31,000 in a month with Anthropic and feels qualified to critique them based on direct experience.

Sponsor Segment: Code Rabbit 01:00

  • Code Rabbit is an AI tool for code review, providing summaries, diagrams, and feedback within pull requests.
  • The tool helps catch bugs, reduces manual review workload for teams, and can be customized with rules for coding standards.

Anthropic’s Business Practices and Attitude 03:22

  • Anthropic is accused of cutting off model access to competitors (e.g., Windinsurf, OpenAI).
  • The company projects a humanitarian image but is seen as contradictory and protective of their "secret sauce."
  • A leaked CEO memo shows internal debate over principles vs. business pragmatism, e.g., considering investment sources.
  • Anthropic is unique among major AI labs in not open-sourcing their coding CLI (Cloud Code) and instead issuing DMCA requests against reverse engineers.
  • Cloud Code was considered internal “secret sauce” and nearly withheld from public release to protect productivity advantages.
  • The company’s early advantage lay in tool-calling capabilities, but competition from open-weight models (GLM 4.5, Kimmy K2) is growing.
  • Other labs, such as Google and OpenAI, have open-sourced their CLIs and accepted outside contributions.
  • Anthropic has not released open-weight models nor reduced prices, unlike peers that have made models and access cheaper or more open over time.

Loss of Competitive Edge 06:18

  • Anthropic’s advantage in tool-calling is diminishing as alternatives catch up or surpass their capabilities.
  • Their models previously excelled at UI tasks (e.g., generating Tailwind components), but new models (Horizon, GPT-4.1) now match or exceed them.
  • Openness in the AI sector (e.g., publishing research, releasing model weights) contrasts with Anthropic’s closed approach.
  • Pricing remains high and discounts are minimal (only 5% with a major upfront spend commitment).

Incidents of Access Restriction and Pettiness 11:05

  • Anthropic aggressively restricted access to Windinsurf even though a relevant acquisition hadn’t happened.
  • The company justifies cutoff by claiming competitors use their models to improve rival products, which the presenter deems illogical.
  • Despite blocking access through official channels, models can still be accessed via third-party providers (e.g., Google Vertex, AWS Bedrock).
  • Such restrictions are seen as petty displays of arrogance and insecurity, targeting competitors and even non-competitor projects.
  • Terms of service prevent users from building competing products, raising concerns about Anthropic's openness to developer ecosystems.

DMCA Takedowns and Open Source Issues 15:36

  • Anthropic has issued multiple DMCA requests to take down reverse-engineered versions of Cloud Code, even for mistakes in publishing source maps.
  • The process involved having old npm package versions removed, which is rare and demonstrates Anthropic’s influence and determination.
  • Open-source alternatives like ADER exist, limiting the practical secrecy or benefit of Anthropic's aggressive behavior.
  • GitHub’s transparency revealed multiple DMCA requests from Anthropic targeting rehosts of Cloud Code.

Pressure from Competition and Lack of Responsiveness 18:02

  • Innovations by other labs (especially OpenAI) have narrowed the technical gap, encouraging developers to migrate.
  • Anthropic’s business edge has been favored by devs, but loyalty is waning as alternatives become more appealing.
  • The company is challenging to work with regarding support and access, with slow communication and difficult negotiations (e.g., for rate limits).
  • The presenter describes better relations and transparency with both OpenAI and Google.

General Conclusions and Final Thoughts 19:29

  • Working with Anthropic produces “weird vibes,” with strong technology overshadowed by problematic business practices and lack of openness.
  • Some excellent researchers and technologies exist within Anthropic, but company management is marked as problematic, even damaging at times.
  • The presenter calls for more community discussion around Anthropic’s practices and voices hope for more openness and improvement from the company.