Supabase Presents
We surveyed over 2,000 startup founders and builders to uncover what’s powering modern startups: their stacks, their go-to-market motion, and their approach to AI.
The ground moved between 2025 and 2026. This report shows where.
Founder and Company
The respondent base got older, more European, more solo, and less self-described-technical. Experienced operators with AI in their pocket are starting companies again.
Solo founders are 61% of respondents — the largest group on the page. Cofounder pairs slipped. The one-person startup is the dominant shape and is still growing.
78% of founders are technical. The 22% who aren’t lean harder on AI code generation than anyone else.
Every age band above 40 has grown by a statistically significant margin. The 22 to 29 cohort shrank by 5%. Seasoned operators are filling the gap, often with AI doing the typing.
The top metros are still the top metros. But AI has flattened the development gap everywhere else. Europe grew 4%; Africa grew 2%. Startups are setting up across Toronto, Chicago, Denver, Austin, across Europe, Asia, and Africa. The playbook no longer requires a specific zip code.
Q&A
Anthropic generation
Claude Code became the most-named must-have dev tool. Claude paid subscriptions nearly doubled. Anthropic overtook OpenAI on the model-provider question. The Anthropic Agent SDK leads SDK adoption.
Cursor dropped 19%. v0, Bolt, and Windsurf each lost 6–9%. VS Code held flat. Antigravity appeared for the first time and took the 4th spot in its debut year.
Q&A
Paid Claude subscriptions sit at 59% of respondents. Paid OpenAI is at 39%. Gemini entered the list at 27%.
Q&A
Anthropic/Claude climbed from 38% to 64%; OpenAI fell from 69% to 51%. Gemini entered at 43%. Hugging Face and custom models lost material share, suggesting fewer teams are running their own inference.
Q&A
AI-written code
62% of startups have more than half their codebase written by AI. 41% are at 76–100%. Only 2% are at zero. Older founders use it more heavily than younger ones, and non-technical founders more than technical ones.
Tech Stack
Supabase gained ground as a primary database, and combined with Postgres, it’s clear what platform startups are betting on. Hyperscalers lost share. And the frontend layer is diversifying fast.
Postgres went from 76% to 82%. Every legacy NoSQL option lost share: MongoDB dropped 5%, MySQL 3%, Firebase 2%. Neon, DynamoDB, and Convex appeared as options for the first time and took small but measurable shares.
Q&A
Three in four respondents who answered the auth question picked Supabase Auth.
Q&A
Supabase held its lead. Vercel extended its lead over AWS by 9%. Cloudflare grew fastest of all, crossing 27% and passing AWS on the way up. Every hyperscaler lost share.
React and Next.js both grew. But the quieter story is that four tools went from effectively zero to real share in 12 months: Expo 10%, TanStack 8%, HTMX 4%, Astro 3%. Native mobile also picked up 3 points.
AI and Agents
Half of respondents are building agents. Multi-agent systems are in production at a quarter of them. MCP adoption crossed 57% in its first year. But the operational layer beneath all of this is missing: most teams do not monitor AI workloads, most have no formal prompt management, and one in three has no eval process.
Agent-building share is statistically flat year over year. The “not sure” cohort shrank, which means undecided builders are making up their minds and shipping. What they automate has shifted: workflow and data analysis climbed; customer support (chat bots) fell off the top spot.
Q&A
Workflow and process automation
Most common AI agent use cases
Three in four agent-builders are already doing multi-agent work. 25% say it is in production. 21% in development. 36% are planning. Only 16% said no.
Prompt management, evaluation, and AI observability all paint the same picture: most teams ship agents without the operator stack underneath them.
A year after the protocol launched, 57% of respondents have already adopted it. Only 14% say they are unfamiliar.
Q&A
Operator tools
Five separate questions show the same pattern this year. CRMs, analytics, observability, dev communities, prompt management. In every one, named vendors lost share, and either “we don’t have one yet” or “custom-built” grew. Startups are not buying operator tools. They are building their own or doing without.
53% of startups with a GTM motion have no formal CRM, up from 43%. Every named CRM lost share: HubSpot, Salesforce, Notion/Airtable, Google Sheets. Build-your-own, or nothing at all.
Q&A
56% still don’t use observability tools. “Custom solution” is the fastest-growing answer, up 2.5%. Datadog and Prometheus both lost share. Sentry kept its lead and grew slightly. The observability market is bifurcating between Sentry for errors and custom dashboards for everything else.
“I don’t track this yet” grew 5%. HubSpot, Salesforce, Mixpanel, Segment all lost share. Custom-built dashboards grew 6%. The pattern repeats: skip the vendor, ship something internal, or don’t track at all.
“No, we haven’t built a community” grew 4 points to 48%. “In progress / planning to” shrank. The “yes, we built one” share is flat at 11%. Dev-community-led marketing has a smaller top-of-funnel this year than last.
Unified backend platform combining auth, edge, database, and queues
Operator tools startups wish existed
Go-To-Market
Founders still do sales themselves. Two in three have never tried paid acquisition. Pricing is settling on tiered feature plans for the first time.
Personal networks remain the top source of initial paying customers. 67% of respondents still haven’t tried paid acquisition at all, up from 62% last year.
Q&A
Dedicated full-time sales hires usually do not arrive until after the tenth employee. Product-led growth as a motion climbed 4 points to half of respondents. “Not sure yet” is shrinking.
For the first time in the survey, startups are picking a pricing shape earlier in their lifecycle, and they are picking the same one. Tiered feature plans went from 23% to 36% of respondents. The “still experimenting” cohort shrank.
Where they show up
Conferences emptied out. Social media lost users across every major platform except TikTok. 1 in 10 respondents now says they have given up on social media entirely. 1 in 3 says they have no online persona at all.
2 in 3 respondents are not attending any industry conference. The “none of the above” cohort jumped 10 points. Google Cloud Next, AWS re:Invent, Microsoft Build, and Y Combinator Demo Day all lost share. Conference-led developer marketing is working for a smaller slice of the market every year.
Outlook
The biggest single movement in the entire survey: “technical complexity” as the largest business challenge fell from 24% to 11%. AI ate the hard parts of shipping. What replaced it: burn out, AI-competition fear, runway anxiety. Optimism is mostly flat. Engineers less so.
The largest year-over-year shift in any single category. Three new challenge options came online: burn out, AI competition, runway anxiety. Together they absorb roughly the same share that used to pick technical complexity. Among 1–10 person teams, burn out has already overtaken technical complexity as the second-biggest challenge.
Q&A
56% say they are optimistic, down 2 points from last year, not statistically significant. Founders are 58% optimistic; non-founders are 49%. The gap widens among engineers and marketers.
Q&A
Supabase is the Postgres development platform. Build your startup with a Postgres database, Authentication, instant APIs, Edge Functions, Realtime subscriptions, Storage, and Vector embeddings.
A special thanks to the following companies for participating in this year's survey.