Your AI strategy starts on a laptop, not in a cloud

Most companies' AI strategy in 2026 reads like this:

→ Pick an enterprise LLM vendor → Sign a multi-year contract → Stand up a "Center of Excellence" team → Roll out cloud copilots to employees → Build internal RAG infrastructure on cloud → Measure adoption with dashboards

This is the IT-led, top-down, cloud-default AI strategy. It's what AI vendors want you to do. It's also one of the slowest, most expensive ways to deploy AI.

There's a faster, cheaper, more grounded path. It starts on a laptop.

The laptop-first AI strategy

The thesis: AI strategy in 2026 should be built from the bottom up, starting with what individual employees can do on their own laptops, not from top-down enterprise contracts.

Here's what that looks like:

Step 1. Pick 3-5 employees in different functions (engineer, marketer, ops, salesperson, support). Give them Ollama + Avery NXR or equivalent local-first tooling. Tell them: "Build agents to handle your operational work. Local. No restrictions."

Step 2. Wait two weeks. Don't manage them. Don't measure them. Don't make them attend the AI Center of Excellence meetings. Let them figure out what works.

Step 3. At two weeks, ask each person to share: what they built, what worked, what didn't, what they'd recommend their team try.

Step 4. Take the patterns that worked. Document them as templates. Roll them out function by function.

Step 5. AS the patterns spread, build the central infrastructure — connector libraries, knowledge bases, audit policies — that supports what the bottom-up adoption has already validated.

This is the opposite of the typical AI rollout. It's also dramatically more effective.

Why bottom-up beats top-down for AI

Discovery happens at the edge. The employee who processes invoices knows what's annoying about invoice processing. The salesperson knows what's painful about pipeline tracking. Central teams trying to design AI workflows for these employees almost always miss the texture.

Adoption follows ownership. Employees who built the agents themselves use them. Employees who got agents handed down from a central team often don't.

Iteration is faster on personal laptops. No procurement. No security review. No infrastructure team. Just install + build + try. The iteration loop is hours, not weeks.

Failure is cheap. If an employee's agent doesn't work, no harm done. If the central CoE's enterprise AI rollout doesn't work, you've spent months and a contract.

What "local" enables that "cloud" doesn't

The bottom-up strategy depends on local-first tooling specifically. Here's why:

No procurement. An employee can install local-first tooling without IT approval, contract negotiation, or budget request. (Most IT policies allow installing free software.)

No data exposure decisions. If the AI processing happens on the employee's laptop, the "what data can we send to the cloud" conversation doesn't trigger. Security stays happy because data doesn't leave.

No cost management. Local models have zero marginal cost. Employees can experiment freely without worrying about a bill.

No central infrastructure dependency. Employees don't need to wait for the central team to provision API keys, configure rate limits, or stand up infrastructure.

The bottom-up strategy works because local-first tooling removes the friction that would normally require central coordination.

What the central team SHOULD do

Bottom-up AI strategy doesn't mean no central role. Central teams should:

→ Provide tooling. Make local-first agent platforms available. Cover the small per-user cost ($29/month per user for Avery NXR Pro is trivial vs. enterprise LLM contracts).

→ Document patterns. As bottom-up adoption discovers what works, central teams should turn the patterns into shareable templates.

→ Build the integrations. Connector libraries to internal systems (HRIS, CRM, support tools) benefit everyone and require centralized work.

→ Govern the edges. When agents start touching customer data or sensitive systems, central governance is appropriate. Audit ledgers help here.

→ Connect to cloud when needed. For the workloads that genuinely need frontier reasoning, the central team can provide BYOK keys or Consult Mode-equivalent escalation paths.

The central role becomes ENABLEMENT, not DICTATION. Employees do the discovery. Central infrastructure supports the discoveries that worked.

Why most companies will get this wrong

The instinct to do top-down AI is strong because:

→ AI feels strategic and important, so it should be CIO-owned → Vendor relationships matter, so let's pick one and commit → Governance is critical, so let's centralize → Employees can't be trusted to manage this themselves

Most of these instincts are wrong for AI specifically.

AI is different from previous IT categories because the per-employee leverage is so high. Empowering individual employees to build agents that handle their own work creates compounding returns that top-down rollouts can't match.

The companies that figure this out first will look back at the AI Center of Excellence model as a 2024-era artifact, like the data warehouse projects of the 2000s.

What this looks like in 2 years

The companies running laptop-first AI strategy will have:

→ Hundreds of small workflow agents owned by individual employees, each handling specific operational work → Central knowledge bases and connector libraries supporting the agents → Audit ledgers across all agent activity → Cost structure that scales linearly with users, not exponentially with usage → Data residency posture that auditors love → Employees who actually USE AI instead of having AI imposed on them

The companies running top-down strategy will have:

→ One or two enterprise AI products with low actual adoption → A Center of Excellence team that produces white papers → Cloud LLM bills that grew faster than headcount → Customer/auditor concerns about data flowing through external clouds → Employees who got bored waiting for IT to enable real AI use cases and now use ChatGPT personal accounts to do their jobs anyway

The strategy gap will be obvious in retrospect.

Start tomorrow

You don't need a strategy meeting to start this. You don't need executive buy-in. You don't need a budget.

Pick one employee. Give them a copy of Avery NXR (free) on their existing laptop. Tell them to build one agent for their most annoying recurring task. Check in in two weeks.

If it works, do it with 4 more employees in different functions.

Six weeks in, you'll have empirical data about what AI strategy actually works in your company. The bottom-up rollout will be ahead of any top-down rollout that started at the same time.

→ avery.software — Free Desktop tier. Start your AI strategy on a laptop, not in a contract.