TLDR Dev 2026-05-20
Gemini 3.5 Flash 🚀, Gemini Spark ⚡️, Karpathy joins Anthropic 💪
Maintainability sensors for coding agents (10 minute read)
AI coding agents use maintainability sensors like linting and type checking to self-correct quality issues, as these tools provide an automated feedback loop for handling things like complex logic and excessive function length. Despite reducing manual overhead, human oversight is still necessary for preventing over-engineered solutions and addressing deeper semantic quality concerns.
Inverting PhotoDNA (5 minute read)
Microsoft's PhotoDNA uses digital signatures to identify illegal imagery based on the claim that these hashes are irreversible, but a machine learning tool named Ribosome successfully inverted them to reconstruct recognizable images. This shows that PhotoDNA signatures leak a lot of image information.
The Cheap, Good, and Fast Paradox Returns (7 minute read)
AI is challenging the long-standing project management paradox that work cannot be cheap, good, and fast simultaneously by accelerating the technical coding phase. Even with these advancements, human resistance to new tools and complex regulatory hurdles like SOC 2 compliance continue to prevent an easier development process.
Dumb Ways for an Open Source Project to Die (12 minute read)
It's common for open source projects to turn into "zombies,” still used but totally abandoned, because of issues like maintainer burnout, funding problems, or old, broken tech. Since these packages are still listed everywhere, they pose a secret but major risk to all the software that depends on them.
Advancing content provenance for a safer, more transparent AI ecosystem (7 minute read)
To make AI media transparent, we're using a multi-layered strategy that includes C2PA metadata and cryptographic signatures so you can easily verify the origin. This system also embeds invisible SynthID watermarks into images for resilience, and a public verification tool is available to check for all these signals.
Anthropic is Preparing for IPO and We Should Be Worried (5 minute read)
Anthropic seems to be getting ready for an IPO by making its access more restrictive, like limiting 'claude -p' and Agent SDK usage, and cutting off third-party tools and competitor APIs. These actions, plus their major $200M DoD deal, show Anthropic is moving away from its public-interest branding toward being a closed, monetization-focused company.
The most important software engineering news in one daily email
Join 450,000 readers for
one daily email