☕️ OpenAI Just Dropped GPT-5.4 — And Your Enterprise Workflow Will Never Be the Same

This was the week AI stopped being a side project and started being the org chart. OpenAI shipped its most ambitious enterprise model ever — a million-token monster with native financial tools baked in. Jack Dorsey fired 4,000 people and told Wall Street the remaining staff would do 2.6x the work. Goldman Sachs said the productivity boom is real but razor-thin. Perplexity cloned a $30,000 Bloomberg Terminal for $200. And a Cloudera study confirmed what most CIOs already suspected: almost nobody's data is actually ready for any of this. Buckle up.

Key Takeaways:

On March 5, OpenAI released GPT-5.4, a new foundation model the company describes as its "most capable and efficient frontier model for professional work" . The launch is not a subtle iteration — it is a deliberate, full-throated push into the enterprise market. The model ships in three versions: a standard release, GPT-5.4 Thinking (a reasoning-optimised variant), and GPT-5.4 Pro for maximum performance, available to Pro and Enterprise plan subscribers. The API version supports a context window of up to 1 million tokens, by far the largest OpenAI has ever offered, enabling the model to process entire codebases, legal contracts, or financial filings in a single pass.

The headline capability improvements are significant. GPT-5.4 is 33% less likely to make errors in individual factual claims compared to its predecessor GPT-5.2, and overall responses are 18% less likely to contain errors . It set record scores on the OSWorld-Verified and WebArena Verified computer-use benchmarks, and scored 83% on OpenAI's GDPval test for knowledge-work tasks. OpenAI also introduced a new system called Tool Search, which allows the model to look up tool definitions on demand rather than loading all of them into the system prompt upfront — a meaningful efficiency gain for enterprise deployments running dozens of integrated tools. Native financial plugins for Microsoft Excel and Google Sheets are included at launch, a clear signal that OpenAI is targeting the finance, legal, and consulting sectors that Anthropic's Claude has been quietly winning.

The timing of the launch is not accidental. OpenAI is playing catch-up in the enterprise segment, where Anthropic's Claude has earned a strong reputation for reliability and instruction-following. GPT-5.4 is OpenAI's most direct answer to that challenge, and it arrives alongside the broader OpenAI Frontier platform — an enterprise agent orchestration layer that lets companies build, deploy, and govern AI agents across their existing software stack . For business leaders evaluating AI strategy, this week's launch resets the competitive baseline. The question is no longer whether to adopt AI for professional work — it is which platform to build on, and how fast to move.

Jack Dorsey's Block laid off approximately 4,000 employees, explicitly attributing the decision to an 18-month internal AI integration programme. CFO Amrita Ahuja confirmed that developer productivity at Block improved by 40% per engineer since adopting AI coding tools, and the company raised its 2026 financial guidance despite the cuts. The implied math is striking: each remaining employee is expected to produce 2.6 times the output of 2025. Block is the clearest real-world case study yet of a major company restructuring around AI productivity — and it will not be the last.

Goldman Sachs research note published this week found "no meaningful relationship between AI and productivity at the economy-wide level" . Only 10% of S&P 500 management teams quantified AI's impact on specific use cases, and just 1% quantified its impact on earnings. However, for companies that have successfully deployed AI in customer support and software development, the median productivity gain is 30%. The takeaway for leaders is not that AI does not work — it is that broad, unfocused AI investment does not work. Targeted deployment in the right functions does.

Perplexity's new multi-model AI agent platform, Perplexity Computer, went viral this week after a demo showed it building a real-time stock analysis terminal — functionally comparable to a Bloomberg Terminal — from a single prompt . Bloomberg charges approximately $30,000 per user per year for its Terminal. Perplexity Computer, available on the Perplexity Max plan at $200/month, orchestrates 19 different AI models including Gemini, Grok, and ChatGPT to handle research, design, coding, and deployment in a single workflow. The demo is a proof of concept, not a replacement — but it signals a new category of AI tool that compresses the build cycle for specialised software from years to minutes.

A joint study from Cloudera and Harvard Business Review Analytic Services found that just 7% of enterprises say their data is completely ready for AI adoption, while 27% say their data is "not very" or "not at all" ready. The finding is a critical counterweight to the week's product launches: the gap between what AI can do and what most organisations can actually deploy is still enormous. Data governance, data quality, and integration infrastructure remain the primary bottleneck — not the models themselves.

The AI hardware race is also heating up. Broadcom CEO Hock Tan announced the company expects to generate “significantly in excess of $100 billion” in AI chip revenue in 2027, a clear shot at Nvidia’s market dominance.

Legendary investor Howard Marks of Oaktree Capital stated on CNBC that AI is poised to eliminate a vast amount of knowledge work, a significant shift from his previous skepticism . His comments add major weight to the argument that AI's economic impact will be widespread and transformative.

Join us on March 13, 2026, at 11:30 AM EST for a three-session event featuring Mark Hinkle, Pam Didner, and Deniz Olcay — covering everything from agentic marketing foundations to measurable content activation to reproducible human-AI workflows that actually deliver results.

Register here →

Keep learning with these upcoming free virtual events from the All Things AI community.

March 10 | The Path to Becoming AI-First: An Operator’s Playbook — This session explores what it really means to become AI-first: starting with process, increasing leverage, and learning how to manage digital employees alongside human employees.

March 17 | Securing your Model Context Protocol Footprint — In this session, Stacklok will demonstrate how to use its open-source ToolHive MCP platform to curate a trusted MCP registry and implement security best practices, including authentication, token exchanges, and access controls—so you can safely deploy MCP-powered agents in production environments.

March 18 | Building Agents with Granite Workshop — We will explore writing AI agents using IBM Granite AI models. We will touch on what are AI agents and explore some different designs for agent development. These designs are informed by the experiences of IBMers in developing agents.

☕️ AI Tangle: Pentagon vs. Anthropic: A $200M AI Contract on the Brink of Collapse

🔮 AI Lesson: I Vibe-Coded a Website From a Voice Memo, and Here's What Happened

🎯 The AI Marketing Advantage: The AI Sales Gap Is Already 77%

💡 AI CIO: Industrializing AI Will Decide the Winners

📚 AIOS: From the 14-day bootcamp to Super-user and Agent Builder programs

This piece from the American Enterprise Institute explains why transformative technologies often look unimpressive in aggregate data before suddenly accelerating. We may be living through that inflection point right now. If you only do one thing this week, find out which two functions in your business — customer support and software development are the most common — are already generating measurable AI productivity gains, and double down there before expanding anywhere else.

All Things AI 2026 — March 23–24 | Durham Convention Center, NC

I produce the All Things AI Conference with my business partner,Todd Lewis, founder of All Things Open. We are committed to upskilling and aim to deliver the most valuable, accessible expert-led workshops in the industry. Here’s what’s on tap in Durham in March. Workshops sold out in 2025. Don't wait. Check out all the workshops here.

Prices increase after March 17. Compare that to $1,000–$3,000+ at other AI conferences.

🎟️ Get Your Tickets Today!