AI Stack by Team Size 2026: What to Buy at Each Stage
The question of which AI tools to buy is always answered wrong by the people who haven't done it before. They either buy too much (enterprise AI platforms they don't have the engineering to configure), or too little (one ChatGPT Plus subscription shared across the company on an honor system). Neither works.
What actually works depends almost entirely on team size. Not because bigger teams have more money, but because the bottlenecks are completely different at each stage. Here's a practical breakdown of what to buy and what to skip, from solo founder to 100-person company.
1-person startup
You're probably a solo founder or a freelancer. You're doing sales, product, marketing, support, and coding either yourself or with heavy AI assistance. Your constraint isn't money, it's time and cognitive load.
Buy:
Claude Pro or ChatGPT Plus ($20/month each). Pick one and actually use it deeply. Don't subscribe to both at first. Claude is better for writing, long-document analysis, and following complex instructions. GPT-4 is better for interactive coding and has a broader tool ecosystem. If you're primarily a developer, start with ChatGPT. If you're primarily a writer or analyst, start with Claude. You can always switch later.
Cursor or Windsurf ($20/month). If you write any code at all, an AI code editor pays for itself in the first hour of use. Cursor's agent mode will write entire features from a description. Windsurf is comparable and has a slightly more aggressive default for auto-accepting changes. Don't use plain VSCode with a Copilot plugin if you're doing serious development work.
Notion AI (included in Notion's paid plans, $12/month). Your notes, documents, and knowledge base are already in Notion. Having AI that can summarize your notes, draft from templates, and search across your workspace is genuinely useful for a solo operator.
Skip:
Enterprise AI platforms (Harvey, Glean, Moveworks). These are built for companies with IT departments, not solo operators. The per-seat minimums and onboarding complexity make them non-starters.
Any AI analytics or BI tool. At one person, you're probably not sitting on enough data or enough time to configure these.
Total monthly spend: $50-$60/month.
5-person team
At five people, you're a seed-stage startup or a small agency. Everyone is doing a mix of things. You probably have one or two engineers, a product or design person, and someone handling GTM. The AI strategy shifts from individual productivity tools to shared resources.
Buy:
Team plans for your core model provider. Either Claude Team (Anthropic) or ChatGPT Team, not both. These give everyone access to the best models, include higher rate limits than individual plans, and don't have conversations feeding into training. At roughly $30/person/month for Claude Team and $25/person/month for ChatGPT Team, budget $125-$150/month for the whole team.
Cursor or Windsurf for every developer. This is non-negotiable. Every engineer on the team should have one. At $20/person/month, two engineers cost $40/month total and they'll each get back several hours per week.
Linear + AI features (included in paid plans). If you're shipping software, Linear for project management with its built-in AI summarization and issue triage is worth it. This is already baked into what a growing team should be paying for, so it's not really an "AI add-on."
An async communication tool with AI. Loom for video messaging, which now transcribes and summarizes automatically, is excellent for async teams. $12-$16/month per person, but you probably only need it for 2-3 people who do a lot of external communication.
Consider:
A lightweight knowledge base with AI search. Notion AI, Outline, or Coda are all reasonable. The use case is making sure the team can find decisions, processes, and documentation without asking each other. At five people this starts to matter.
Skip:
Dedicated AI agents for specific functions (recruiting, marketing). Too early. The time cost of configuring and maintaining these exceeds the value at this scale.
Total monthly spend: $300-$400/month.
20-person team
Series A or later. You have functional departments, even if small ones. Engineering, sales, marketing, customer success, maybe HR. The AI tooling decision is now partly a procurement decision, not just a personal productivity choice. Someone needs to own it.
Buy:
One model provider with an enterprise-adjacent plan for the whole company. Both Anthropic and OpenAI have team/business plans with central billing and admin controls. Standardize on one. Splitting the company across model providers creates unnecessary tool fragmentation without meaningful benefit. Budget $25-$30/person/month.
GitHub Copilot Business for all developers. GitHub Copilot Business ($19/developer/month) includes company-managed licenses, policy controls, and doesn't use your code to train the model. If you have 5 developers, that's $95/month well spent.
A GTM AI layer. At 20 people you probably have a sales team. Clay ($200-$800/month depending on tier) for prospect research and enrichment combined with an AI writing assistant for outreach (Lavender, Amplemarket, or similar) meaningfully improves outbound throughput. Expect to spend $400-$600/month total on this function.
Customer support tooling with AI. If you have any customer-facing support volume, Intercom or Zendesk with their AI triage and response drafting features are worth activating. This is a capability built into tools you might already be paying for. If you're not using it, turn it on.
Loom for internal and customer communication. At 20 people with remote or hybrid work, async video becomes important. AI transcription and summarization is now standard in Loom's paid plans.
Consider:
An AI meeting assistant (Fireflies, Otter.ai, Granola). These join your meetings, transcribe, summarize, and extract action items. At 20 people with multiple meeting threads, the value of searchable meeting archives grows quickly. About $10-$20/person/month.
Cursor or Windsurf for all engineers (not just some). Consistency matters. If half the engineering team has AI code editors and half doesn't, you get inconsistent practices and a harder time maintaining shared context.
Skip:
Dedicated AI HR tools, AI legal tools, AI finance tools. At 20 people, these are over-specialized. You don't have enough volume in any single domain to justify the category-specific tooling.
Total monthly spend: $2,000-$3,500/month depending on engineering headcount and sales team size.
100-person company
Series B or beyond. You have real departments, compliance requirements, probably some enterprise customers with vendor requirements, and a budget review process. AI tooling is now a line item in the annual plan, not a month-to-month experiment.
Buy:
Microsoft 365 Copilot or Google Workspace AI. At this scale, you're almost certainly on one of these platforms already. Activating AI across the productivity suite (Word, Excel, Outlook, Teams or Google Docs, Sheets, Gmail, Meet) for most employees is cost-effective and reduces the number of point solutions you need to manage. Microsoft 365 Copilot adds $30/user/month on top of existing M365 licenses. Google Workspace AI is included at Business Plus and above. For 100 people, this is a $2,500-$3,000/month line item that covers a lot of surface area.
GitHub Copilot Enterprise for all developers. At this scale, the enterprise tier ($39/developer/month) adds custom model fine-tuning on your codebase, deeper integration with your internal code, and admin controls for audit logs. If you have 20 developers, that's $780/month for meaningfully better code completion that knows your internal APIs.
A dedicated AI enablement for sales. At 100 people you probably have 15-30 people in sales and marketing. Clay, Apollo, or Outreach with AI features for prospecting; an AI writing tool for email sequences; and Gong or Salesloft with AI call analysis. Budget $3,000-$6,000/month for the full GTM AI stack.
An enterprise AI search tool. Glean or Elastic's enterprise search with AI gives employees a way to search across all internal knowledge (code, docs, tickets, Slack, email) using natural language. At 100 people with 2-3 years of accumulated documents and conversations, this becomes genuinely valuable. Budget $5,000-$8,000/month.
Observability and governance. You need to know what AI tools employees are using, what data is being sent to external services, and whether any of it violates your enterprise customer agreements. Tools like Nightfall for DLP, or simply AI usage policies enforced through an MDM, become necessary. This is partly a process cost and partly a tooling cost.
Skip or defer:
Custom-built internal AI agents for non-core workflows. At 100 people you have the engineering capacity, but many teams overbuild here before validating that the specific agent delivers ROI. Build agents for your highest-volume, most-standardized workflows first.
Total monthly spend: $25,000-$45,000/month depending on how many departments are covered and the depth of the sales AI stack.
The pattern that cuts across all four stages
The most common mistake I see is buying tools prematurely. A 5-person team buying enterprise search. A 20-person company buying custom agent platforms before they've standardized on a base model provider. The tooling rarely delivers value before the team is large enough and structured enough to actually use it.
The second most common mistake is not standardizing. Companies that let every individual choose their own AI tools end up with no shared knowledge base, no consistent practices, and much higher total spend than they realize. At 5 people, individual choice is fine. By 20 people, someone needs to make a call and enforce it.
The third mistake is ignoring the operational cost of AI tools. Every additional platform means another integration to maintain, another contract to renew, another vendor relationship to manage. Each tool should either replace something you're already paying for, or deliver measurably more value than its monthly cost. If you can't articulate what a tool is saving you or enabling, it probably shouldn't be in the stack.
Buy less, use it better, and add tools as you outgrow what you have.