The Small-Team AI Stack, 2026 Edition: What Five People Actually Need

Every small team now has the meeting: “what AI should we be paying for?” The vendor answer is everything; the honest answer is a short stack, adopted in the right order. Here’s the 2026 playbook we give five-to-twenty-person teams, with budget math and the two popular purchases that don’t survive contact with reality.

Layer one: the assistant standard (do this first)

Pick ONE general assistant as the team default and buy seats deliberately. The differences among the leaders matter less than the standardization: shared prompts, shared expectations, one tool in the onboarding doc (the current head-to-head if you’re choosing). Budget: seats for the heavy users, free tiers for occasional ones — most teams overbuy here by half. The compounding win is cultural: when everyone shares a tool, the good prompts and workflows spread hallway-fast.

Layer two: the coding lane (if you ship software)

The highest-ROI seat in the stack, and the one where per-seat stinginess costs real money. The 2026 choice is workflow-shaped — ambient assistance versus agentic editing (our daily-use verdict) — and the honest rollout is a two-week trial where each developer tracks hours saved. Approve whatever the tracking justifies; this line item routinely returns 5-10x its cost, and the broader case is made in our coding-tools cost-benefit analysis.

Layer three: the research seat (one, not many)

One cited-research subscription (the Perplexity-vs-Gemini decision) for whoever writes the memos, competitive analyses, and client briefs. This is a role purchase, not a team purchase — one or two seats, shared outputs. Teams deep in Google Workspace should test whether Gemini’s bundled tier covers the need before adding a standalone.

Layer four: images, by evidence

If last month’s design requests were logo-adjacent and social-graphic-shaped, DALL-E inside the assistant you already bought covers it free-marginally. Buy the dedicated image subscription (Midjourney, honestly reviewed) only when a creator role generates weekly at volume — then buy it happily, it’s the category’s best money for that human.

The two purchases to skip (for now)

The all-in-one “AI workspace” platform promising every capability in one dashboard — in practice a worse version of each layer above at a bundle price; and per-seat AI on every SaaS you already rent (the +$10/user/month toggles multiplying across your stack) — audit which ones anyone opened twice before renewal. The discipline: capabilities are cheap, sprawl is expensive.

The budget math, assembled

A representative ten-person team: assistant seats for six, coding seats for three, one research seat, zero dedicated image (yet) — roughly $250-350/month, replacing scattered individual spending that typically exceeded it. Run the subscription audit quarterly, and let the free tiers’ rise cancel lines for you. The stack that wins isn’t the biggest — it’s the one your team actually opens every morning. Tool-by-tool current rankings: the directory.