Sora's Gone. Your AI Now Needs a Token Budget.

OpenAI just shut down Sora, the AI video app that shook the media industry two years ago. The underlying model is genuinely impressive-it generates video and audio at a level that still unnerves people-but there was no sustained product-market fit. The Disney partnership is ending too. It's a reminder that technical capability and actual user traction are not the same thing.

AI Access Is Becoming a Recruitment Tool

More interesting than any single product shutdown: how AI is reshaping compensation. Microsoft's Charles Lamanna revealed that job candidates are now negotiating AI token budgets as part of their offer. We're talking $100 to hundreds of dollars per day-essentially a new line item in the fully-loaded cost of an engineer. If an engineer costs $500k annually and asks for $100k in tokens to be 3x more efficient, that's a good deal for everyone. Nvidia's Jensen Huang agrees-tokens are becoming a recruiting tool in Silicon Valley. This isn't speculation. It's happening now.

Your AI Agents Still Have Goldfish Memory

Here's the harder problem: most AI agents still don't know what just happened. A user cancels a flight, but the AI agent is still looking at yesterday's data. The gap between transactional systems (where reality lives) and analytical systems (where AI looks) is a latency disaster. Aurora Zero-ETL and Redshift Vector Search fix this-near-zero latency between your database and AI memory. The old Glue pipeline approach took 30+ minutes. This takes seconds. For AI agents to actually feel intelligent, they need to know the present moment, not yesterday's batch.

Web Tools That Actually Save Time

On the builder front, we've got two comprehensive guides that cut through the noise. Code formatters and linters are the lowest-effort, highest-ROI investment in code quality-Prettier, ESLint, and the new Biome (25-35x faster than Prettier on large files) are the real options in 2026. Secrets management has evolved from emailing .env files to real solutions like Doppler and Infisical, both of which handle the sync problem that haunts small teams.

The pattern here is about removing friction from the work itself. Token budgets, live memory systems, automated formatting-these are all the same conversation: what can we automate so humans focus on what actually matters. Sora's failure doesn't change that direction. If anything, it sharpens the focus on tools that actually get used.