One-Line Summary

OpenAI moves deeper into day-to-day enterprise and consumer money workflows with GPT-5.5 and Codex, while chip-market and courtroom narratives put cost and trust at the center of AI adoption.

Big Tech

OpenAI expands GPT-5.5 into enterprise agents and personal finance

OpenAI and Databricks are putting GPT‑5.5 to work on real company tasks and giving ChatGPT a finance view that connects to bank and investment accounts. Databricks says GPT‑5.5 is now available for customer agent workflows after it set a new high on OfficeQA Pro, its benchmark for complex enterprise document tasks, cutting errors by 46% versus GPT‑5.4 and becoming the first model to surpass 50% accuracy. These gains show up on parsing-heavy jobs like scanned PDFs and legacy files that often break production agent systems. ¹

OpenAI is also releasing a preview of a new personal finance experience in ChatGPT to Pro users in the U.S., letting them securely connect accounts (via Plaid) across more than 12,000 financial institutions and ask questions grounded in their real financial context. OpenAI frames this as a measured rollout to learn from early use, and notes that ChatGPT is not a replacement for professional financial advice. The preview is available on web and iOS for Pro users in the U.S. ²

In parallel, Sea Limited, a Singapore-founded tech group behind large e-commerce and digital finance operations, outlines how it is rolling out Codex to help developers navigate big codebases and CI/CD pipelines with agentic workflows. Internal data cited by Sea shows 87% of users are weekly active, and among developers rating Codex 4 or 5 out of 5, 73% would recommend it, pointing to sustained usage beyond autocomplete. ³

These pushes into sensitive workflows arrive as TechCrunch highlights that jurors are weighing trust in OpenAI’s leadership after closing arguments by May 17, 2026, with a broader media discussion about how little visibility outsiders have into private AI labs. The pairing of new finance features with renewed scrutiny underscores why privacy, data control, and verifiable reliability matter as adoption deepens. ⁴

Industry & Biz

AI chip rally raises cyclicality concerns

The Wall Street Journal argues that investors may be underestimating how the chip industry’s boom‑bust cycles still apply, even amid strong AI demand, with memory makers a particularly extreme case. The analysis cautions that “it’s different this time” narratives can overlook how quickly supply, pricing, and profits swing. ⁵

As one example, WSJ notes Micron recorded its biggest-ever loss three years ago but is now forecast to become the sixth‑most profitable U.S. stock, projected to make just under $100 billion over the next 12 months—more than Meta or Berkshire Hathaway—which illustrates both optimism and potential vulnerability to cycle turns. ⁵

Cerebras recounts near-failure before $60B debut

TechCrunch details how AI chip maker Cerebras, now worth about $60 billion after a blockbuster IPO, nearly failed in 2019 while burning about $8 million a month and “incinerating nearly $200 million” trying to solve packaging for its wafer‑scale chip. The team ultimately fixed cooling and data‑movement challenges—at one point inventing a machine to bolt 40 screws simultaneously to secure the wafer—before the system finally worked. ⁶

The report also says OpenAI loaned Cerebras $1 billion secured by warrants for roughly 33 million shares (worth over $9 billion at a $279 closing price), and that Cerebras agreed to a time‑limited restriction on selling to specific OpenAI competitors as part of that deal. Today, Cerebras sells AI chips for inference to customers including OpenAI and AWS. ⁶

What This Means for You

Improvements on OfficeQA Pro suggest fewer brittle edges in document-heavy workflows; if your team wrangles scanned PDFs or legacy files, ask your vendor whether they can show GPT‑5.5 performance and error deltas versus prior models in an agent-harness setup before you commit. ¹

For personal money tasks, the ChatGPT finance preview can centralize accounts and questions in one place, but it’s U.S. Pro‑only for now, links via Plaid, supports 12,000+ institutions, and is not professional advice; review ChatGPT’s Data controls so your settings match your comfort with training. ²

Engineering and product leads can study Sea’s shift from autocomplete to agentic workflows inside CI/CD, with usage data (87% weekly active; 73% of highly satisfied users would recommend) hinting at durable value in code understanding, debugging, and test generation. ³

Procurement and finance teams should pair chip‑cycle caution with capacity realities: the WSJ’s cyclicality warning and reporting on Cerebras’s customer‑prioritization history both argue for explicit SLAs, capacity guarantees, and price‑protection clauses in AI infrastructure and model contracts. ⁶

Action Items

1. **Try ChatGPT’s finance preview ** (U.S. Pro): Open the Finances panel in ChatGPT, connect accounts via the guided flow, and review Settings > Data controls to confirm your training preferences.
2. Pilot a GPT‑5.5 agent on real documents: If your team uses Databricks, ask an admin to enable GPT‑5.5 in AI Unity Gateway and run a 1–2 hour test on scanned PDFs to compare error rates with your current setup.
3. Run a 90‑minute “agentic workflow” workshop: Map three recurring engineering or operations tasks to AI agents, define success metrics (latency, error rate), and pick one pilot to implement this month.
4. Tighten AI vendor contracts: Ask IT/procurement to add explicit capacity SLAs, price‑adjustment protections, and exit clauses to upcoming AI infrastructure or model agreements.