# How startups are bundling software and AI tools The startup tech stack is experiencing its biggest shakeup in years. See why AI tools like Anthropic, OpenAI, and Cursor are reshaping how startups build. **URL Source:** https://www.brex.com/journal/brex-benchmark-startup-software-stack --- The Brex Benchmark How startups are bundling tech: 2020 vs 2025 _The Brex Benchmark explores how today’s companies spend on AI and SaaS tools, as well as other business essentials. We analyzed the spending patterns of our startup customers to understand how their tech stacks have evolved over the past five years. The findings reveal fascinating patterns in AI tool adoption and what this means for the emerging startup infrastructure. (Startups are defined as companies with fewer than 250 employees.)_ **The key insight:** AI is reshaping the startup tech stack, with startups leaning into AI tools wherever they can optimize existing workflows. They are diversifying tools when switching costs are low (models like **OpenAI** and **Anthropic**), and standardizing on tools when switching costs are high or there is tighter integration (like **Cursor** for development and **Figma** for design). **What we found:** - **Adobe** is out, **Figma** is in for design tooling. - **OpenAI **is now essential startup infrastructure. - **Notion** looks to be challenging **Google Workspace**. - Startups are building core dev workflows around **Cursor**. - Voice-first bundles are forming around **ElevenLabs** and **Deepgram.** - Right now, it’s **Instantly** and **Clay** for sales and marketing automation. ![OpenAI, Figma, and Cursor are essential startup software.](https://brand.brex.com/transform/90dcadae-f949-4eb8-a1d9-08e7c1675eb4/benchmark-startup-tech-stacks-july2025) The startup stack evolution In 2020, shortly after Brex launched what is now the [Brex business account](https://www.brex.com/product/business-account), the five most common startup products bundled on Brex from 2020-21 were Adobe, Amazon (AWS), Google, Slack, and Zoom. By 2023, Figma began competing with Adobe head-to-head, becoming the second-most common design tool. Fast forward to 2025, and there are two key changes to the top 5: - **Figma** is the design tool of choice for startups, with **Adobe** no longer appearing in even the top 10 bundles. Since 2020, Figma's market share jumped by 47% while Adobe's dropped by 38%. This marks a turning point in how design gets done, from static files to seamless design-to-prototype. Figma leads that future, with **Cursor**, **Replit**, and **Framer** buyers 2x more likely to choose Figma over Adobe. Top teams are betting on browser-based, collaborative tools to drive modern design workflows. - **OpenAI** is among the most common bundles for startups in 2025, despite **Anthropic** being the leader in total dollars spent, according to our [May 2025 Brex Benchmark](https://www.brex.com/journal/brex-benchmark-may-2025). **Google, Slack, **and **Amazon (AWS)** continue to maintain a strong presence in the startup stack. ![July brex benchmark startup evolution](https://brand.brex.com/transform/549bec3b-72b8-47ad-94bb-83971c61ebd4/July-brex-benchmark-startup-evolution) Claude: The AI model startups scale with **OpenAI** is the default for many early-stage startups. But as they mature, especially those building AI-first products, they increasingly scale spend with **Anthropic**, suggesting Claude is their foundational model for powering core product experiences. - **Anthropic **customers rarely purchase Claude alone. Despite leading in [total AI spend](https://www.brex.com/journal/brex-benchmark-may-2025), most adopt it as part of a multi-model strategy, using Anthropic to round out and strengthen their AI stacks. - **Together.ai** runs open-source AI models like Llama and Mistral, and it’s often positioned as an **Anthropic** and **OpenAI** alternative. However, 67% of Together.ai customers purchase Anthropic, suggesting Claude is worth the premium for specific use cases. - 78% of **Anthropic** customers also buy **OpenAI**, but only 20% of OpenAI customers reciprocate. Notion vs. Google Workspace One interesting race to watch is in the workspace collaboration market. **Notion**'s growing feature set has expanded beyond its database-wiki roots into email and productivity tools, including Notion Mail. Yet 67% of Notion customers still pay for **Google Workspace**, suggesting Notion remains more complementary than competitor. However, the trajectory feels familiar: a specialized tool gradually expanding its scope to challenge incumbents, much like **Figma**'s path before **Adobe**'s 2022 acquisition attempt (and Figma’s planned 2025 IPO). Notion's ability to carve out workspace real estate while customers maintain dual subscriptions suggests we're watching the early stages of a productivity software shift. Cursor is a core development tool **Cursor **has become essential startup infrastructure, ranking as a top-3 AI tool according to Brex Benchmark data over the past few months. The emergence of the AI-powered development environment mirrors how Figma became standard for design and Slack for communication. The top tools startups are pairing with Cursor: - Customers purchasing **Replit **for vibe coding, **incident.io **for incident response**,** or **Pave** for compensation management are 20x more likely to also use **Cursor.** - 43% of customers using **Graphite**, an AI-powered code review tool, also use **Cursor**. While Cursor accelerates code writing, Graphite optimizes the code review and merge process. Also, side question: Are the PMs vibe coding?** **We found that startups paying for [Shreyas Doshi's courses](https://shreyasdoshi.com/courses/), which are designed for product managers, are 20x more likely to bundle them with **Cursor**. How startups are approaching voice-first products [Voice AI apps](https://www.brex.com/journal/brex-benchmark-june-2025) were everywhere in our June 2025 Brex Benchmark. Today, we're seeing specialized bundling patterns emerge around core tools like **ElevenLabs, Deepgram, Speechify,** and **Assembly.ai.** We’ll be curious if our future data shows more infrastructure patterns, since voice AI requires significant computational resources. Customers purchasing GPU compute through **Runpod** are 2x more likely to also use **ElevenLabs** for AI voice generation. - Startups that purchase **Assembly.ai’s** voice-to-text capability are buying **Intercom**’s customer service solutions 43% of the time. - When customers purchase one voice AI tool, they're 3x more likely to bundle it with **Deepgram**, suggesting some solution testing before standardizing. Building the AI-powered sales and marketing funnel Startups are using AI to accelerate customer engagement and reach new audiences. Core startup tools in our data show **Instantly** for cold email outreach and **Clay HQ** for enriching prospect data and outbound automation. - Startups using **Instantly** are 2x more likely to bundle it with **RB2B Pro** for visitor tracking, allowing them to reach prospects and understand their engagement patterns. - 54% of customers who buy **La Growth Machine** for automated outreach campaigns also buy **Clay HQ**. - Other popular tools bundled with **Clay HQ**: - **RB2B Pro** (44%) to identify visitors - **Phantombuster** (33%) to find contact information - **Smartlead.ai** (49%) to automate outreach Advice for startups, and looking ahead Successful startups treat AI as infrastructure, not experimentation. They're building integrated workflow stacks where tools amplify each other, and this bundling trend is accelerating. **Advice for startups:** Start with a core AI tool for your primary workflow, then systematically add complementary tools that enhance the entire process. For example, you might adopt **Cursor** for AI coding, then layer in **Graphite** for code review, **GitHub** for version control, and **incident.io** for operations. This creates a cohesive AI-powered development workflow rather than scattered point solutions that create friction. Looking ahead,** **we expect bundling patterns to solidify around these core AI tools as startups move from experimentation to standardization. The winners will be those building **integrated AI stacks, not AI collections**. **What else we're watching:** - **AI-first categories:** Just as Cursor transformed code editing, we expect AI-native tools to emerge in customer support, sales, and operations, creating new bundling clusters. - **More Figma moments:** Just as Figma displaced Adobe, we're watching for similar disruptions in productivity (Notion vs. Google), collaboration, and project management. _Sumeet Marwaha is the Senior Manager of Data at Brex, supporting the team in understanding how Brex customers spend, adopt tools, and grow their businesses._ _All analysis conducted for this report that uses Brex internal customer data is anonymized and aggregated for privacy. To learn more about how we use data in anonymized or aggregated form for these trend reports, email us at privacy@brex.com._ ## Related Articles ### [Building autonomous agents for technical tasks: 5 lessons learned](https://www.brex.com/journal/building-autonomous-agents-for-technical-tasks) Brex built a general-purpose template for delegating engineering work to agents with confidence. Here's how we did it and what we learned. ### [📋 Controller's Council Webinar](https://www.brex.com/resources/controllers-council-using-ai-register) Brex's Chief Accounting Officer explores how AI can automate manual tasks, deliver outputs for contract reviews and flux analysis, and free up time for growth. ### [Forrester survey article: Better compete globally](https://www.brex.com/journal/forrester-global-spend-management) Top finance leaders share their challenges with global spend management and why the time to solve them is now. ### [July 2025 - Brex Benchmark](https://www.brex.com/journal/brex-benchmark-july-2025) Startups are locking in AI capacity before demand explodes. Should enterprises follow suit? See the latest in The Brex Benchmark. ### [June 2025 - Brex Benchmark](https://www.brex.com/journal/brex-benchmark-june-2025) See which tools startups and enterprises are using to build AI voice agents in our June 2025 Brex Benchmark. ### [How we test our agents: by committing fraud](https://www.brex.com/journal/articles/simulation-testing-ai-audit-agent) Simulation testing lets you find out how good your agent actually is before the world shows you. Here's what we've learned in teaching agents to catch fraud.