Adam Mosseri says Instagram has 'reined in' AI costs after shutting down 'the silly things' that were burning tokens
THE SO WHAT
When a major consumer app talks about killing “silly” AI features to rein in token burn, it’s a reminder that AI UX is now a cost center with hard constraints. Teams should be instrumenting feature-level AI spend and pruning low-value prompts long before finance forces caps.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIEnterprise AI is entering an evaluation gap: Agents are gaining autonomy faster than companies can verify them
If half of enterprises have shipped agents that passed evals yet still failed in front of customers, your test harness is now a production risk, not a compliance checkbox. Treat agent autonomy like you’d treat a new payments flow—add staged rollout, kill switches, and human-in-the-loop gates before you widen scope.
Applied AI'Today, the token economy is emerging': AI use in China increases 1000-fold with usage exceeding 140 trillion tokens a day
140 trillion tokens per day and 1,000x growth in China means AI economics are converging on metered compute, not seats or MAUs. If you sell or buy AI, you need a tokens-per-outcome view of unit economics this quarter, or your pricing model will be misaligned with how usage is actually scaling.
Applied AIPrivacy advocates, CAA, and SAG-AFTRA slam Meta's Muse Image, which lets users create AI images using the likenesses of people with public Instagram accounts
If public Instagram photos are treated as fair game for likeness generation, every brand, creator, and employee with a public profile just picked up a new deepfake risk surface. Map where your people’s faces are exposed and tighten social media, consent, and takedown playbooks before this becomes a crisis-management problem.
Applied AIMonitoring systemic drift may guide the next phase of organizational resilience
As AI links more systems together, the real failure mode becomes slow, invisible drift in dependencies and behavior. Treat AI observability like SRE for socio-technical systems — map cross-domain dependencies and set alerts on behavior changes, not just uptime.