BREAKING: Perplexity Just Split… | The AI Corner Summary

1. Key Themes

The "unpasted folder" is the real frontier for AI adoption, not model capability

The article's central thesis is that the biggest bottleneck for AI usage isn't intelligence, it's trust with sensitive data. This reframes the AI race from "who has the smartest model" to "who can touch the data that matters."

"AI adoption stalls at exactly one place: the folder you refuse to paste."

Hybrid (split cloud/local) architecture is a genuine product strategy, not just a privacy feature

Perplexity's bet is that splitting a single task across two models — one for reasoning, one for privacy — unlocks previously unautomatable workflows rather than just being a compliance checkbox.

"Privacy pitches usually ask you to accept a weaker tool for safer handling. This one gives the agent access to the folder you had been keeping away from it."

On-device compute economics change the cost structure of agentic work

Running inference locally isn't just about privacy — it has a direct cost advantage that scales with how much of a task can be pushed off the cloud.

"the local half burns zero cloud credits. Work running on your Mac costs nothing per token, so the more of a task that stays on-device, the cheaper the whole thing runs."

Apple legitimized the hybrid-compute pattern, and Perplexity is extending it into agentic territory

The article frames Apple's Private Cloud Compute as proof-of-concept and Perplexity's move as the harder, more valuable next step (multi-step file-based agents vs. single prompts).

"Apple's version handles single prompts. Perplexity handles agentic tasks that read files and take multi-step actions, where the agent opens your documents, extracts from them, and ships artefacts."

Growth data suggests enterprise demand is shifting toward privacy-preserving agentic tools

Perplexity's revenue trajectory is used as evidence that the market is rewarding this architectural bet, not punishing it for being slower/costlier to build.

"The company shipping Hybrid Compute is running at $750 million in annualized revenue, per Sacra's estimates, up from $232 million at the end of 2025."

2. Contrarian Perspectives

Building the harder, slower architecture can be a competitive advantage, not a liability. Conventional wisdom favors shipping fast with a pure cloud model; the article argues the fastest-growing player deliberately chose the harder path because it saw where enterprise demand was heading.

"A hybrid architecture is more expensive to build and slower to ship than a pure cloud agent. The fastest-growing player in AI search chose to build it anyway... I read that as a bet on where the enterprise money is heading: toward the folders nobody will upload."

Trained classifiers beat regex-based privacy filters because context matters more than pattern-matching. Most privacy tooling is dismissed as security theater built on brittle pattern matching, while Perplexity's contextual classifier is framed as a meaningfully different (and better) approach.

"A regex can match a shape. It can't read the situation."

The published methodology matters more than the unreleased benchmark — verify the claim yourself rather than wait for third-party validation. The author explicitly acknowledges the evidence gap yet still endorses trying it, offering a DIY test instead of waiting for proof.

"So the method is documented today, while the benchmark that would let you test it independently is still coming."

3. Companies Identified

Perplexity

Description: AI search/agent company, maker of Perplexity Computer and the new Hybrid Compute feature.
Why mentioned: Central case study of the newsletter — the company shipping the hybrid cloud/local architecture and Privacy Gate classifier.
Quote: "Hybrid Compute on Perplexity Computer runs one task across 2 models... Computer decides which model sees each step, then stitches the two halves into one answer."

Apple

Description: Maker of macOS/Apple silicon hardware and Private Cloud Compute.
Why mentioned: Cited as the company that "made hybrid respectable," providing the architectural precedent Perplexity is extending.
Quote: "Apple made hybrid respectable. Private Cloud Compute set the pattern: do what you can on-device, escalate what you must to a cloud you can verify, and make the boundary explicit."

Sacra

Description: Research/data firm providing startup revenue estimates.
Why mentioned: Source of the revenue growth figures used to substantiate Perplexity's momentum.
Quote: "running at $750 million in annualized revenue, per Sacra's estimates, up from $232 million at the end of 2025."

Google (Drive), Gmail

Description: Productivity/data storage tools referenced as connectors.
Why mentioned: Used as integration points in the deal-room workflow example.
Quote: "Connectors: Drive ingests the data room locally. Gmail surfaces banker email without it leaving the device."

Granola / Claude (referenced in "Keep reading" links)

Description: Tools referenced in a related past article about private context stacks.
Why mentioned: Cross-referenced as complementary approach to private AI workflows.
Quote: "Build a private context stack with Granola and Claude"

4. People Identified

Ruben Dominguez

Description: Author/writer of the newsletter piece.
Why mentioned: Byline for the article; frames the narrative through personal experience with the "redact, paste, un-redact" workaround.
Quote: "I know the workaround because I ran it for a year. Redact, paste, un-redact, repeat."

(Note: No other named individuals — e.g., executives, researchers — are mentioned in the article; the PII-Tracer paper's authors are referenced only anonymously.)

5. Operating Insights

Design prompts around the model's actual strengths, not aspirational ones. Since local models are smaller than frontier cloud models, operators should architect workflows so reasoning-heavy work stays in the cloud and only extraction/comparison tasks run locally.

"Design your prompts so the heavy reasoning runs on public context and the local side does extraction, filling, and comparison."

Reduce onboarding friction as a growth lever — the "one-click" install matters more than model quality for adoption. The article credits a single UX decision (in-app local model download) with solving the adoption barrier that kept an entire category niche.

"Step 2 is the one that kept local models a hobbyist thing until now. The friction was never the model. It was the 45 minutes before you could ask it anything."

Enterprise-readiness requires policy infrastructure, not just capability. For team/enterprise sales, admin-level configurability and auditability are what unlock procurement approval, more than the underlying AI feature itself.

"If you are evaluating this for a team rather than yourself, the relevant part is that enterprise admins set the rules org-wide... Audit logs sit behind those decisions."

6. Overlooked Insights

Remote-triggering of local compute (iPhone → home Mac) quietly creates an always-on private agent infrastructure, which is a meaningful architectural pattern beyond the privacy narrative — essentially turning a home Mac mini into private inference infrastructure.

"you can fire a task from your iPhone while the Mac at home runs the local half. Point it at an always-on Mac mini and you have a private agent that works whether or not you are at the desk."

Platform fragmentation is an underappreciated constraint: Hybrid Compute launches Mac-only, with Linux (DGX Spark) already available and Windows still pending, meaning a large share of enterprise (Windows-dominant) users are currently excluded.

"Linux builders already have a Perplexity local agent (Portable Computer, on DGX Spark), and Windows lands in September."