AI for business in 2026: what changed and what matters for decision-makers

Updated October 2026. Everything here has a date and a source. When something is a forecast, I say whose.

1. Agents: from pilot to production, without the rush

An agent is AI that doesn't just answer: it carries out a sequence of steps, like reading an order, checking inventory, drafting the quote and asking for your sign-off. Agents got much better in 2026. Stanford's 2026 AI Index, released in April, shows agent success on a test of real computer tasks (OSWorld) jumping from about 12% to 66% in one year. Humans score 72% on the same test.

But almost nobody has scaled them. In McKinsey's State of AI survey from November 2025, 23% of organizations said they were scaling agents somewhere and another 39% were only experimenting. Gartner's 2026 Hype Cycle for agentic AI is blunter: only 17% of organizations have deployed agents, though more than 60% expect to within two years. Vendors are seeing real demand: Salesforce reported on August 26, 2026 that Agentforce annual recurring revenue topped $1.5 billion. Note the company's own caveat: that quarter the metric started including other AI products.

Why it matters for a midsize business

Agents can now do real work, but they still fail about a third of the tasks on that test. For a decision-maker that means: an agent running alone, with no rules and no review, not yet. An agent with a tightly scoped task and human approval on the decisions that matter, yes.

What to do

• Pick one repetitive task with a clear start, middle and end (for example, triaging orders that arrive by email and text).
• Decide what the agent can do on its own and what has to come to you.
• To see how several agents work together, read AI agent orchestration.

2. MCP and A2A: AI's plugs became standards

Think of a wall outlet. Before there was a standard, every appliance had its own plug. AI was the same: connecting a model to your spreadsheet, CRM or inbox took a custom integration for every pair. Two protocols fixed that. MCP (Model Context Protocol) connects AI to tools and data. A2A (Agent2Agent) lets one agent talk to another, even across vendors.

On December 9, 2025, Anthropic donated MCP to the newly formed Agentic AI Foundation under the Linux Foundation, with OpenAI and Block as co-founders and support from Google, Microsoft and AWS. By then there were more than 10,000 active public MCP servers, and ChatGPT, Gemini and Microsoft Copilot already used it. A2A, created by Google, reached a stable 1.0 release in March 2026 and, per the Linux Foundation, had more than 150 supporting organizations by April. On August 17, 2026, it became a project of the same foundation as MCP.

Why it matters for a midsize business

Open standards lower the risk of vendor lock-in. If your tools speak MCP, you can switch AI models without rebuilding integrations. It's the same idea behind an AI operating system: your knowledge and rules stay with you, and the model becomes a swappable part.

What to do

• Ask every software vendor: “Do you have an MCP connector?” A vendor with no plan for one deserves a hard look at renewal.
• Don't pay for a custom integration that a standard connector already handles.

3. The price of intelligence collapsed (the bill, less so)

A token is the unit AI is billed in, roughly a piece of a word. On September 22, 2026, the research group Epoch AI estimated that the cost of reaching a given level of performance has fallen about 13x per year since 2023. A concrete example: on November 24, 2025, Anthropic launched its strongest model at $5 per million input tokens and $25 per million output tokens, a third of the prior generation's price.

Epoch flags its own caveats: the estimate assumes users always pick the cheapest model that can do the job, and benchmarks don't capture everything that matters at work. Agents also burn far more tokens per task than a chat. The price per unit falls, but the number of units rises.

Why it matters for a midsize business

Cost is no longer the main barrier for most office tasks. The barrier now is having a process, organized data and clear rules. And any long contract tied to a single model ages fast.

What to do

• Avoid multi-year contracts locked to one specific model.
• Ask vendors for a monthly spending cap and a usage report.
• Use smaller, cheaper models for simple tasks; save the frontier model for work that needs reasoning.

4. AI moved into the tools you already pay for

Since January 2025, Google has included Gemini in Workspace business plans, with a price increase (annual Business Standard went from $12 to $14 per user). Microsoft answered with a small-business Copilot at $21 per user per month starting December 1, 2025, for companies with up to 300 people. On April 22, 2026, agent features in Word, Excel and PowerPoint became generally available to subscribers, and on May 1 Microsoft launched Agent 365, a console to inventory and govern agents, at $15 per user per month.

Google went further on October 8, 2026, announcing a work agent inside Gemini Enterprise that can run tasks for hours or days, in private preview for a small group of customers. Google says there will be no extra charge for existing Gemini Enterprise customers but hasn't detailed usage limits yet.

Why it matters for a midsize business

You're probably already paying for AI you don't use. At the same time, each tool sees only its own slice: Copilot sees your documents, the CRM sees the pipeline, and nothing sees the whole company. Built-in AI speeds up tasks, but it doesn't pull together the information that's scattered across group chats, spreadsheets, email and people's heads.

What to do

• List the AI licenses you already pay for and how many people actually use them.
• Turn on what's already included before buying anything new.
• Write down which decisions still require pulling information from three places. That's where a Business AIOS makes a difference.

Find out how much of your company still depends on you.Take the free index

5. Regulation: US in a tug-of-war, Europe delayed, Brazil pending

United States

On December 11, 2025, the White House signed Executive Order 14365, which created a task force to challenge state AI laws. The most visible case is Colorado: according to an analysis by law firm Squire Patton Boggs, enforcement of the state's AI Act was stayed by a federal court on April 27, 2026, after a lawsuit by xAI that the Justice Department joined. Expect more of this back-and-forth before there's a clear national rule.

European Union

Per the European Commission's official timeline, the AI Act's bans and staff AI-literacy duty have applied since February 2, 2025, and rules for general-purpose models since August 2, 2025. On August 2, 2026, the law became broadly applicable, including transparency rules such as telling people when they're talking to an AI. The Omnibus package, agreed on May 7 and in force since July 27, 2026, delayed the high-risk rules: to December 2, 2027 for areas like hiring and education, and to August 2, 2028 for AI built into regulated products.

Brazil

If you do business in Brazil: its AI bill, PL 2338/2023, passed the Senate on December 10, 2024 and is still awaiting the rapporteur's report in a special committee of the lower house. On August 24, 2026, the rapporteur said the vote should come at year end, after the October elections. That's his expectation, not a scheduled date.

What to do

• Don't wait for new laws: existing privacy and consumer-protection rules already apply to any personal data that flows through AI.
• Tell customers when they're talking to an agent. It's good practice and already the rule in Europe.
• If you sell into Europe or operate in several states, ask counsel for a map of obligations by jurisdiction.

6. High adoption, rare returns: what 2026 research shows

Almost everyone uses AI. Few make money from it. In PwC's global survey of 4,454 CEOs, released in January 2026, 56% said they had seen no significant financial benefit from AI, and only 12% reported both higher revenue and lower costs. Among US small businesses, Goldman Sachs found in March 2026 that 76% use AI, but only 14% have fully embedded it in core operations. In Brazil, IBGE found AI use among industrial companies with 100+ employees rose from 16.9% in 2022 to 41.9% in 2024.

Why it matters for a midsize business

PwC notes that CEOs seeing gains on both sides are the ones who embedded AI broadly across the business, on top of solid data and governance. In practice: the model isn't the bottleneck. The bottleneck is AI that isn't connected to how the company actually runs.

How to read news without losing focus

The pace of AI news creates a trap: the impression that you have to react to all of it. Almost never. Three questions filter most of the noise before the team’s time is spent.

1. Does this change a decision the company will make in the coming months? If not, it is information, not action.
2. Does it require data, process or permission we do not yet have? If so, the first job is to prepare the base, not to adopt the novelty.
3. Can it be tested small, with a deadline and a measure? If not, it is still early.

A habit that helps: set a fixed time each quarter to look at the changes that matter, with one person responsible for bringing the summary. The rest can wait at no cost.

Decision checklist for next quarter

You don't need to track every launch. You need to make five or six good decisions. This is the list I recommend for the next 90 days.

Your next step

What changed in 2026 is that technology stopped being the limit. The limit now is clarity: knowing where your company loses time, which decisions only you make and which information lives scattered around. Before choosing any tool, it's worth measuring that.

A good start is the Dependency Index (IDD), a free quiz that shows on a 0 to 100 scale how much the company depends on you. To understand the idea of bringing everything into one system, read what an AIOS is and see how it works, one area at a time.

Cesar Augusto Carlos

Founder, NoOne Systems

Has been building several businesses with digital marketing since 2008. Today he designs systems that bring data, processes and AI together so owners can see the whole company and decide what matters.