Microsoft Build 2026 is about to kick off, and if the early messaging is anything to go by, we can expect a continued focus on AI agents, Microsoft 365 Copilot, Azure AI Foundry, GitHub Copilot, and Microsoft’s vision of what it calls the “Frontier Firm”.
For those who have not been following the terminology, Microsoft’s Frontier Firm concept describes organisations that combine human workers with AI agents to create highly flexible teams built around outcomes rather than traditional organisational structures. According to Microsoft’s recent Work Trend Index, business leaders increasingly expect AI agents to become part of mainstream operations over the next 12 to 18 months. The vision is compelling: humans focus on judgement, leadership and creativity while AI handles routine execution.
As a long-time observer of technology trends, I suspect Build 2026 will double down on this message.
I only hope Microsoft and the broader AI industry dial down the hype slightly and spend more time discussing the practical realities.
The first issue is cost.
At present, many of the most capable AI models feel as though they have been priced for venture-funded startups and enterprise innovation budgets rather than independent developers, consultants, researchers, or small businesses.
The promise of agentic AI sounds attractive until somebody receives the invoice.
One of the uncomfortable realities of API-based AI is that costs are often difficult to predict upfront. A poorly designed workflow, an unexpected loop, or a runaway prompt can consume substantial resources before anyone notices. In traditional software projects, costs are generally bounded and predictable. In contrast, many AI workflows operate more like a taxi meter running in the background.
For businesses trying to build repeatable processes, this creates a significant challenge.
Who wants to commit an important business process to a technology where the final cost may only become apparent after execution, and where success is never guaranteed?
The AI vendors need to address this issue quickly. Cost uncertainty is rapidly becoming one of the biggest barriers to adoption. If organisations cannot estimate operational costs with confidence, many promising AI projects will remain experimental rather than moving into production.
My own experience over the weekend reinforced this concern.
I decided to experiment with OpenAI Codex and asked it to help generate a relatively simple five-slide PowerPoint presentation. The excitement lasted until I reviewed the API costs. The exercise cost roughly the same as a decent restaurant meal.
That may not sound significant in isolation, but multiply that by dozens of experiments, prototypes and iterations, and the economics quickly become difficult to justify.
The second issue is branding.
Microsoft desperately needs to simplify its Copilot portfolio.
Today we have Microsoft 365 Copilot, GitHub Copilot, Copilot Studio, Security Copilot, Copilot Chat, Copilot-enabled applications, and a growing collection of AI experiences carrying the same label.
Ask ten people what “Copilot” means and you may receive ten different answers.
A strong brand should create clarity. At the moment, the Copilot brand just creates confusion.
Personally, I would not object if Microsoft completely rebranded the entire portfolio tomorrow, provided the result made it obvious which products are enterprise-grade, which are developer-focused, and which are entry-level offerings. Customers should not need a scorecard to understand what they are buying.
Perhaps the most interesting consequence of current pricing trends is the renewed interest in open-source self-hosted AI.
Over the weekend I found myself researching self-hosted models, local inference options, and open-source alternatives. Not because I necessarily believe they outperform the leading frontier models, but because experimentation needs to remain affordable.
There currently appears to be a large gap in the market. The cheapest models are often fast but unreliable. The frontier models are remarkably capable but expensive. There seems to be surprisingly little middle ground.
That is a pity because many independent developers, consultants and entrepreneurs would happily trade a small reduction in performance for dramatically lower costs.
Despite these reservations, I remain enthusiastic about Microsoft Build.
The conference consistently provides valuable insight into where the industry is heading, and Microsoft’s engineering teams continue to produce genuinely impressive technology. I will be following the sessions closely and soaking up the information like a sponge.
My hope is that alongside the excitement about AI agents and Frontier Firms, we also hear more discussion about practical deployment, realistic economics, governance, and the limitations of current systems.
And that’s a wrap for today. My OpenAI account has decided to deduct another $5 from my credit card for researching this blog.
