Our Shoe Company Is Now An AI Company

Snap is eliminating approximately 1,000 full-time employees—roughly 16% of its global workforce—while also closing more than 300 open roles. The company projects the combined actions will reduce its annualized cost base by more than $500 million by the second half of 2026. CEO Evan Spiegel cited AI-driven productivity improvements as a central justification, framing the cuts as enabling teams to "reduce repetitive work, increase velocity, and better support community partners and advertisers." The layoffs arrive under compounding pressure. Snap's stock has declined approximately 31% year-to-date, weighed down by regulatory challenges to teen social media access, mixed results from an advertising business overhaul, and an unproven augmented reality glasses initiative. Activist investor Arenik Capital Management, which recently acquired a stake and publicly called for workforce reductions, adds a governance dimension to what management is framing as an AI-driven operational decision. The company reported first-quarter revenue of approximately $1.53 billion, a 12% increase, with adjusted EBITDA (earnings before interest, taxes, depreciation, and amortization) of roughly $233 million. Whether AI efficiency or investor pressure is the primary driver remains an open question the memo did not resolve.

The discussion covers one of the more dramatic corporate reinventions in recent memory. Allbirds—a San Francisco-based footwear brand once valued at over $4 billion—was sold this month for $39 million, representing a decline of more than 99% from its 2021 Nasdaq flotation price. The acquiring entity, American Exchange Group, will continue operating the shoe brand and serving existing customers. What remains is the public company shell, traded under the ticker BIR, which the new leadership intends to repurpose as an AI compute infrastructure provider under the proposed name Newbird AI. The company plans to raise $40 million via convertible notes from an undisclosed institutional investor and use the proceeds to acquire GPU assets—graphics processing units used to power AI workloads—offered to customers as a cloud service. The strategy is explicitly compared in the discussion to the 2017 Long Island Ice Tea Company pivot to blockchain, which produced a brief 275% stock surge before the company was delisted the following year. Stockholder approval is required, with a meeting scheduled for May 18.

OpenAI is deploying GPT-5.4 Cyber, a fine-tuned variant of its GPT-5.4 model, to a limited set of vetted participants in its Trusted Access for Cyber program—a framework launched in February 2026 to verify the identities of cybersecurity professionals. The model is designed for defensive cybersecurity applications, specifically identifying software vulnerabilities so organizations can remediate them. Notably, it operates with a lower "refusal boundary" than standard models, meaning it is more permissive in allowing users to probe its capabilities for legitimate security research purposes. Initial access is limited to hundreds of users, with plans to expand to thousands in coming weeks. OpenAI began integrating cyber-specific safeguards into its deployments in 2025 and launched a separate vulnerability-identification tool, Codex Security, in March 2026.

Google has released a Windows desktop application—available on Windows 10 and above, in English globally—that functions similarly to macOS Spotlight, allowing users to search the web, Google Drive, local files, and installed applications from a single interface triggered by a keyboard shortcut. The app incorporates AI Mode, Google Lens integration, and screen-sharing capabilities for contextual queries. A macOS version is described as in development by the Gemini team. Separately, Google has introduced "Skills" within the Chrome browser's Gemini sidebar—a feature allowing users to save and reuse structured AI prompts via a forward-slash shortcut. More than 50 preset skills are available, covering tasks such as summarizing YouTube videos and recipe modification. Users can also build custom skills containing multiple sequential instructions. The discussion notes that Microsoft and Perplexity AI have deployed comparable features in their respective browsers, while OpenAI's Atlas browser does not yet support the format.

A counterintuitive pattern is emerging in legal services: AI tools used by clients are generating more work for law firms, not less. Partners at multiple UK and US firms describe receiving high volumes of AI-generated emails, patent applications, litigation strategy suggestions, and draft correspondence—much of which requires significant attorney time to review, validate, and correct before it can be acted upon. One US litigation partner described a client producing so many AI-generated emails that the firm established a policy of responding only at intervals and only to material points. A patent attorney noted that AI-generated patent filings frequently arrive at her firm after encountering problems at the patent office, requiring remediation. For fixed-fee clients, firms are absorbing these review costs; for hourly clients, the time is simply billed. One senior partner observed that AI-generated letters often fail to conform to a firm's house style, creating additional revision work. The overall assessment from multiple practitioners is that AI is not reducing legal workload in aggregate—it is redistributing and in some cases amplifying it. Key takeaways: - AI efficiency is becoming a standard public justification for workforce reductions, but the degree to which it reflects genuine productivity gains versus cost-cutting pressure from investors remains difficult to disentangle from external cases like Snap's. - The Allbirds-to-Newbird pivot illustrates how public company shells are being treated as vehicles for AI sector entry, echoing speculative rebranding patterns seen during prior technology cycles. - OpenAI's tiered, identity-verified access model for cybersecurity AI represents a deliberate attempt to manage dual-use risk while expanding capability access to vetted professionals. - Google's desktop app and Chrome Skills reflect a broader platform strategy to embed AI interaction into operating system-level workflows, intensifying competition with Microsoft's Copilot integration. - Law firms are experiencing AI-generated client output as a net workload increase, suggesting that productivity gains from AI tools may accrue unevenly—benefiting the party generating content while imposing review costs on the party receiving it.