Install sentence_transformers and Python tools hang for several extra seconds on cold start — Python 3.15 ships this week, and the official team has finally made "lazy loading" native syntax via PEP 810, but overall performance matches 3.14. We read this as a release where developers win on startup speed and lose on performance expectations.
What this is
Python 3.15 is led by release manager Hugo van Kemenade. This update concentrates on engineering efficiency, not execution speed:
1. PEP 810 "Explicit Lazy Imports": use lazy import json instead of import json; modules load only when a symbol is first referenced. The most direct value for AI developers: many plugins depend on heavy packages like torch and sentence_transformers, which used to add several wasted seconds to cold start for command-line tools like llm.
2. PEP 798 "Unpacking in Comprehensions": you can now write [*sublist for sublist in lists], removing a layer of looping — an extension of PEP 448 from 2015.
3. Tachyon, a sampling profiler built into the standard library: periodically samples call stacks of running Python processes to locate slow functions. Third-party tools like py-spy used to be required; now it ships in the standard library, useful for diagnosing bottlenecks in production AI inference services.
On performance, Miguel Grinberg's testing concludes it is "essentially the same" as 3.14.
Industry view
Noted AI developer Simon Willison publicly called PEP 810 "the one I'm most excited about" — for a direct reason: his own llm command-line tool slows down when heavy-dependency plugins are installed. Willison carries weight in the AI engineer community, and his judgment will shape upgrade priorities for a cohort of tool authors.
But beyond the positive voices, criticism exists too. Performance is Python's chronic pain point: the existence of Mojo, Pyodide, and various Rust-accelerated Python toolchain projects is itself a market expression of dissatisfaction with CPython's speed ceiling. 3.15 leaving performance untouched means no official answer on that front in the near term.
Another layer of controversy is the standard library's continued expansion: folding a sampling profiler into the official tree is good for enterprise stability, but it spreads maintainer attention thin, and complaints about PEP review cadence slowing down have been steady for years.
Impact on regular people
For enterprise IT: Python remains the default language for AI/ML with no way around it. 3.15's features will surface in the next generation of AI products (command-line tools, agent frameworks, IDE plugins) through faster startup and better observability — upgrade planning can begin.
For individual careers: Direct beneficiaries are developers building AI tools and plugins. Non-technical roles will wait 6–12 months until mainstream agent frameworks and IDEs actually migrate to 3.15 before feeling any difference.
For the consumer market: This release is effectively invisible to end users — only when AI desktop tools and agent applications begin requiring 3.15+ will ordinary users encounter it indirectly.