Bluesky’s new app is an AI for customizing your feed

Bluesky’s team launched Attie, an AI assistant powered by Anthropic’s Claude and built on the AT Protocol, that allows users to create custom natural-language feeds and — eventually — generate code for apps on the protocol. Currently in closed beta with a waitlist (attie.ai), Attie could broaden the AT Protocol ecosystem and lower barriers to app creation, boosting user engagement over time, though near-term market impact is limited.

Analysis

Agent-driven personalization on open application layers is a structural amplifier for compute and identity demand rather than a one-off product bump. If even 5–10% of social or content engagement migrates from monolithic apps into many specialized agent-driven feeds over 12–36 months, cloud/accelerator spend could rise materially while ad inventory fragments into long-tail, high-intent slices that require new measurement and bidding stacks.

The immediate second-order winners are commoditized infrastructure providers (GPU, cloud, vector indexing, identity/auth) because developer-onboarding friction falls and more apps mean more API calls and model inferences. Expect marginal cost per engaged user to move from cents to dollars as LLM-driven personalization becomes interactive — at scale a 10M-DAU app adding $1–3/month in inference spend implies $10M–$30M/month in incremental total addressable cloud spend before monetization.

The main risk is monetization and moderation: long-tail feeds create granularity but also make RPMs volatile and moderation costs per impression rise. Regulatory and content-liability regimes could force platform-level controls that dampen the openness advantage, potentially reversing user migration in 6–24 months if compliance costs spike.

Competitive dynamics: incumbents can blunt disruption by acquiring composability tooling or exposing their own agent endpoints; conversely, a thriving open-agent ecosystem makes many small targets attractive M&A candidates and raises takeover value for middleware specialists. Monitor developer activity, API billings, and moderation-cost metrics as leading indicators of durable adoption.

Key Decisions for Investors

• Long NVDA (6–12 months): buy a modest call position or call spread to capture a 30–60% upside if agent workloads materially lift GPU demand; downside limited to premium paid. Catalyst: quarterly evidence of higher inference hours/enterprise contracts from AI-native apps. Risk: architectural substitutes or pricing falls in GPUs.
• Overweight MSFT (12 months): accumulate shares or buy long-dated calls as a play on Azure capturing higher-margin LLM inference and identity/graph integrations. Expect 15–30% upside if commercial AI infra revenue growth accelerates; risk is slower enterprise migration or competitive pricing from AWS/Google.
• Long OKTA (9–18 months): buy calls or shares to play rising demand for federated identity and cross-app auth as open-agent apps proliferate. Upside 25–50% if OEMs and multi-app ecosystems standardize on third-party auth; downside is margin pressure and execution risk.
• Pair: short META vs long NVDA or MSFT (9–12 months): initiate a defensive pair (e.g., buy NVDA calls / sell META put spread) to express view that feed commoditization will pressure ad RPMs while infrastructure vendors benefit. Target 2:1 asymmetric payoffs — limited premium outlay for outsized upside if ad fragmentation materializes; main reversal risk is rapid incumbent feature imitation and retention of ad monetization.