Raising the Baseline

Anthropic released Claude Sonnet 5 in June 2026, offering improved performance and lower pricing compared to the November 2025 Opus 4.5 model, with input costs dropping from $5.00 to $2.00 per million tokens. The author argues that while AI has made software production easier, human coordination and operational challenges remain the bottleneck for building reliable systems at scale.

Raising the Baseline Table of Contents I’ve been writing about AI Coding Assistants , LLMs, Agents , and Learning in the Agentic Era . It might be relevant for this post –– or eventually outdated. This week I was talking with family members about AI, one medical doctor and one commercial director. None of them had developed software before. All of them are now solving a lot of their professional papercuts with software produced by AI. The baseline has been raised Producing software is easier than before. November 2025’ Opus 4.5 https://www.anthropic.com/news/claude-opus-4-5 moment is now cheaper with Sonnet 5 https://www.anthropic.com/news/claude-sonnet-5 . We’re still navigating the exponential improvements on models’ capabilities and benefiting from the current AI scaling laws https://blogs.nvidia.com/blog/ai-scaling-laws/ . | Model | Launch | Input $/MTok | Output $/MTok | Context | Max Output | |---|---|---|---|---|---| | Claude Opus 4.5 | Nov 2025 | $5.00 | $25.00 | 200K | 64K | | Claude Sonnet 5 | Jun 2026 | $2.00 intro → $3.00† | $10.00 intro → $15.00† | 1M | 128K | † Sonnet 5 introductory pricing applies through August 31, 2026. claude-api skill and fetching Models overview https://platform.claude.com/docs/en/about-claude/models/overview page. Don’t get trapped. This is not about producing software for the sake of producing software . It’s all about programmable outcomes at unprecedented speed . Most of them through software solutions . You can’t avoid the hard part Everything beyond a prototype or a “selfish software” is still constrained by human and operational laws: build the right thing at large settings is still ambiguous . People still have opinions. A lot of people will have a lot of opinions. Communication, while increasingly enhanced by AI tooling, is still hard. Synchronization across humans is the bottleneck . Building things right is still hard. Operational safety, security, availability, and reliability are still high-judgement engineering practices . The heavy-lifting is being increasingly reduced by AI tools like AWS DevOps Agent https://aws.amazon.com/devops-agent/ , Datadog’s Watchdog https://www.datadoghq.com/product/platform/watchdog/ and Bits AI Agents https://www.datadoghq.com/product/ai/bits-ai-agents/ . It’s expected that major capabilities will keep getting better for the foreseeable future although nobody can accurately predict 12 months+ ahead in terms of AI advancements . Professional teams were taught and optimized to manage the hardest parts. Several times, that meant avoiding those completely. You can’t avoid the hard part https://xcancel.com/i/article/2073902681668931927 . In abundant times /posts/2026/07/05/the-renaissance-of-abundance , frontier teams are actively seeking out the hardest parts to start working with first . Well equipped with AI teammates. Back to basics Fundamentals are still fundamental to a sustainable pace. If you’re a net new vibe coder and also a cardiologist or if you’re learning single-page applications design patterns along with Netlify deployment options through Claude Code sessions while being a medical imaging equipment seller during the day , welcome to our industry Second, ask Claude or the agent you’re used to to gently teach you about modern development practices and operational safety: Gently teach me about modern software development lifecycle as if I was a your regular profession here . Focus on safety, data security, reliability, and maintainability. Don’t assume any prior experience in the software engineering field. What a time to be building 🚀