Something shifted this week. Not in a headlines-only way, but in a way that changes the practical calculus for any advice firm that has been cautiously experimenting with AI tools and wondering when the real decisions start. They have started.

On 2 and 3 June, OpenAI and Microsoft both made significant announcements about where enterprise AI is going. The direction of travel is consistent: we are moving from what the AI Daily Brief called the “subsidy era” (abundant, cheap compute, exploratory use) into a “scarcity era” (constrained infrastructure, rising costs, and pressure to deploy carefully rather than broadly).[1] For firms in professional services, that distinction matters more than it might first appear.

What the shift from subsidy to scarcity actually means

The practical implication is that AI is becoming infrastructure, not a trial. The decisions you make now about which tools to deploy, on which workflows, at what cost, are beginning to look more like capital allocation decisions than experiments.

SK Hynix, one of the primary manufacturers of the high-bandwidth memory that AI servers require, has announced plans to double manufacturing capacity by the end of the decade. Its chairman has stated that chip shortages could persist until 2030.[2] That is not a cyclical blip. It is a structural constraint on the availability and cost of AI compute for the foreseeable future. For firms choosing between tools or scaling agentic workflows, cost per task is becoming the relevant metric, not cost per query.

The Microsoft cost argument

Mustafa Suleyman, Microsoft’s CEO, made a pointed claim at Microsoft Build this week: when Microsoft’s MAI model was fine-tuned for McKinsey’s workflows, it delivered higher quality outputs than GPT-5.5 on those specific tasks, at ten times lower inference cost.[3]

This is worth taking seriously, not because McKinsey’s workflows resemble yours (they almost certainly do not), but because the principle translates. A general-purpose frontier model is expensive to run at volume. A model tuned to a specific, well-defined task can match or exceed its quality at a fraction of the cost. For advice firms with standardised workflows (annual review letters, meeting summaries, research summaries, suitability note drafts pending human review), that cost gap is likely to become a meaningful line in your technology budget within eighteen months.

The implication is not that you need to start fine-tuning models yourself. It is that the vendor you choose, and whether they offer workflow-specific optimisation rather than raw frontier model access, is now a more important selection criterion than it was six months ago.

The knowledge work problem OpenAI is naming

OpenAI published a report this week titled “The Next Era of Knowledge Work,” which includes McKinsey data showing the average knowledge worker spends more than 25% of their week on email and around 20% searching for internal information.[4] OpenAI frames this as a “strange abundance” problem: technology made producing documents, dashboards, and messages cheap, which multiplied the volume of artifacts, and shifted the bottleneck from creation to consumption, coordination, and sign-off.

That description will be recognisable to anyone running a firm above about ten people. The compliance file that nobody can find. The trail of email threads that contain the actual decision. The paraplanner who spends Friday afternoon chasing approvals rather than drafting.

OpenAI’s position is that Codex, its knowledge work platform, addresses this by reducing coordination friction rather than simply generating more content. Codex has five million weekly active users, and knowledge workers (non-developers) currently make up 20% of that base and are growing three times faster than the developer segment.[2] That is a meaningful signal: these tools are no longer primarily for technical teams.

“Knowledge work is still waiting for its factory redesign.”, OpenAI, The Next Era of Knowledge Work[4]

The parallel task shift

One of the more striking behavioural findings from OpenAI’s data: 50% of Codex users now run multiple tasks simultaneously, up from fewer than 33% in mid-April 2026.[2] Users are running dataset reviews, script drafts, and report assembly in parallel, operating as orchestrators of AI tasks rather than sequential executors of individual ones.

This is not just a feature update. It represents a shift in how people are relating to these tools. The mental model moves from “ask the AI a question” to “assign the AI several tasks and review the outputs.” For a senior adviser or practice manager, the comparison is less to a calculator and more to a capable junior colleague who can work on multiple things at once, provided you review what comes back before it goes anywhere.

That last part is not optional in a regulated context. Any AI or automated output that touches a regulated document, a client communication, or a compliance decision requires human review before use. The parallel task model does not change that. It changes how much you can sensibly supervise.

What the regulatory context adds

The weekly picture across UK financial services regulation is not standing still. The FCA, Bank of England, and HM Treasury issued a joint statement in May 2026 confirming that frontier AI models now exceed baseline cyber resilience capabilities, moving AI governance from internal best practice toward a formal regulatory obligation for UK financial services firms.[5]

This is the context in which these OpenAI and Microsoft announcements land for your firm. The tools are maturing rapidly. The cost structure is shifting. The regulatory expectations are hardening. Those three things together mean that “we’re still figuring out our AI strategy” is a less defensible position than it was a year ago.

CrowdStrike’s 2026 Global Threat Report documents an 89% year-on-year increase in attacks by AI-enabled adversaries.[6] The same AI capability that makes your internal workflows more efficient is being used to make external attacks more sophisticated. The firms that treat AI purely as a productivity tool, without considering what it means for their security posture and governance obligations, are accumulating a risk they may not yet have fully priced.

The regulatory signal from the US

The Trump administration issued an executive order on AI earlier this year that reduced the pre-release notification window for frontier AI models and kept safety testing voluntary, with the stated intention of maintaining innovation pace.[7] Some observers have argued that the classified review thresholds and vague “meaningful step change” criteria are, in practice, laying the infrastructure for a future model licensing regime, regardless of the current framing.[1]

This is a US policy story, not a UK one. But it matters here for two reasons. First, the AI tools you are using are predominantly US-developed and US-regulated at the infrastructure layer. Policy decisions that constrain or shape those development cycles affect what you can deploy and when. Second, the direction of travel in AI governance is broadly consistent across jurisdictions: regulators are moving from hands-off to structured oversight. The FCA’s joint statement is the UK expression of the same trend.

What this means in practice for a firm your size

The source material from this week’s announcements suggests three practical questions worth putting to your team in the next month.

First, map your AI use against cost. If you are using AI tools, you are likely paying per query or per seat at general-purpose frontier model rates. Identify which workflows run most frequently and whether a more cost-optimised approach (a different tool, a tuned model, a simpler automation) could handle those tasks at lower cost and equivalent quality. You do not need to do this for everything. Start with the highest-volume, most standardised workflows.

Second, separate creation from coordination. OpenAI’s “strange abundance” framing is useful. If AI tools are helping your team produce more documents but the time spent reviewing, approving, and routing those documents has not fallen, you have addressed the wrong bottleneck. The question worth asking is not “can AI draft this?” but “where is the actual friction in our process, and is that where the AI sits?”

Third, review your governance posture. If you are running AI tools in any client-facing or compliance-adjacent workflow, you need a clear, documented answer to three questions: what does the AI produce, who reviews it before it is used, and what is the audit trail? That is not a complex project. It is a requirement that is moving from good practice to expected standard.

None of this requires a large investment or a formal AI strategy document. It requires honest answers to practical questions, and a willingness to act on them before the regulatory conversation catches up with your current setup.

If you want to think through what AI deployment and workflow automation could look like specifically for your firm, a discovery call with Cordrey Consulting is a good place to start.

This article is for informational purposes only and does not constitute regulated financial advice or a compliance opinion. Consult a qualified compliance professional for advice specific to your firm.

Sources

[1] AI Daily Brief, ‘The Next Wave of Enterprise AI’ (June 2026). Podcast episode.

[2] Bloomberg, SK Hynix Plans to Double Capacity to Easy Memory Chip Crunch (June 2026). URL: https://www.bloomberg.com/news/articles/2026-06-02/sk-hynix-to-double-wafer-capacity-to-ease-memory-chip-crunch

[3] Microsoft, Microsoft AI CEO unveils 7 new AI models | Mustafa Suleyman at Microsoft Build 2026 (June 2026) Youtube Video. URL: https://www.youtube.com/watch?v=OvLIae4HCeM

[4] OpenAI, The Next Era of Knowledge Work (2026). Cited for knowledge worker share of Codex users (20%), growth rate (3x faster than developers), and the statistic that 50% of Codex users now run multiple tasks simultaneously. URL: https://cdn.openai.com/pdf/the-next-era-of-knowledge-work.pdf

[5] FCA, Bank of England, and HM Treasury, AI and Cyber Resilience: Joint Statement (May 2026). Cited for confirmation that frontier AI models now exceed baseline cyber resilience capabilities and that AI governance is a formal regulatory obligation for UK financial services firms. URL: https://www.fca.org.uk/news/statements/fca-boe-treasury-joint-statement-frontier-ai-models-cyber-resilience

[6] CrowdStrike, 2026 Global Threat Report (2026). Cited for 89% year-on-year increase in AI-enabled adversary attacks. URL: https://www.crowdstrike.com/global-threat-report/

[7] The White House, Executive Order on Artificial Intelligence (2026). Cited for reduction of the pre-release notification window for frontier AI models and voluntary safety testing provisions. URL: https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/