Two stories today are really the same story: AI is reshaping what it means to do technical work, and nobody has settled on the vocabulary yet. Riot Games is apparently porting legacy code in ways that weren't possible 18 months ago, and an Anthropic engineer is proposing a whole new map of job roles to replace the ones that are blurring. Meanwhile, Box CEO Aaron Levie is back on the security beat, this time with a sharper argument about why gatekeeping AI models doesn't actually make anyone safer.
01
The job title obituary nobody asked for, but everyone needed
Boris Cherny, who works on the Claude Code team at Anthropic, posted a thread that's gotten over 11,000 likes and counting. His argument: as engineering, design, product, and data science collapse into each other, new archetypes are emerging to replace old job titles. He identifies five on his team: the Prototyper (generates ideas, most don't ship), the Builder (turns prototypes into production), the Sweeper (cleans up, simplifies, removes), the Grower (scales what's working), and a fifth he cuts off in the preview. The thread is being passed around because it names something people are feeling but struggling to articulate. ---
Why it matters: Your company's next reorg is going to use org chart categories designed in 2015. This framework is more honest about what people actually do when AI handles the routine parts of every discipline. If you're a manager, this is a better map for who to hire and why. If you're an individual contributor, it's a useful question to ask yourself: which archetype are you, and is that what your team actually needs?
Riot Games may be rewriting legacy code with AI agents, and someone noticed
Thariq posted a pointed observation: Riot Games appears to be doing something significant with legacy codebase work, and his hypothesis is that coding agents have changed the math on what's feasible. Porting or modernizing a legacy codebase used to be the kind of project that got budgeted for years and then quietly shelved. If agents can make that tractable, it's not a productivity improvement. It's a category shift in what software projects are worth attempting. ---
Why it matters: Every company sitting on a 15-year-old codebase they can't afford to touch just had their cost estimate invalidated. The question is whether this is real and repeatable, or one team's favorable conditions. Worth watching for follow-up confirmation from anyone at Riot.
Aaron Levie makes the case that AI export controls are security theater
Box CEO Aaron Levie posted a direct argument: advanced cybersecurity models at the level of Mythos will inevitably become open and available to anyone. His conclusion is that restricting US model releases doesn't make the world safer. It just guarantees that the alternative tech stacks that emerge around open models will be built outside American influence. The gatekeeping buys nothing while ceding the architecture of the next decade. ---
Why it matters: This connects to what Guillermo Rauch raised yesterday about offensive AI capabilities. Levie is pushing the argument further: if the dangerous capabilities are coming regardless, the question isn't whether to allow them but who builds the infrastructure around them. That's a meaningful reframe for the policy debate, and it's the kind of argument that tends to age either very well or very badly.
Microsoft builds AI that actually remembers your project from last month
Microsoft Research published a paper at ICML 2026 on Memora, a memory system for AI agents. The core problem it solves: every AI conversation currently starts from scratch. Memora separates what gets stored (detailed memories) from how the AI looks things up (lightweight tags), so agents can remember months of context without having to re-read everything each session. It outperforms existing memory approaches and uses up to 98% fewer context tokens. Code is available on GitHub. ---
Why it matters: The "your AI forgot everything" problem is the biggest practical friction for anyone using AI on multi-week projects. If this approach holds up outside the lab, it removes the main reason you still have to brief your AI assistant from scratch every Monday morning.
Peter Yang posted exactly what a significant portion of the enterprise software market is thinking: Salesforce has been marketing Agentforce for over a year, and it still isn't clear what it does. 45 likes and 18 replies suggests the confusion is widely shared.