Free tools Windows power users keep installed
One-click scans. No signup required.
Startups were buying AI mainly as practical software, not wholesale replacements for employees. The a16z/Mercury analysis, published October 2, 2025, reported spending spread across general assistants, coding and app-building tools, creative software, customer support, sales, recruiting and specialized workflows. OpenAI ranked first and Anthropic second, but the broader result was fragmentation: many products were being purchased for specific jobs.
That conclusion needs a boundary. The analysis covers Mercury transaction activity from June through August 2025, not every startup or every AI dollar. It is best read as a behavioral snapshot of application-layer adoption, rather than a complete measure of the AI economy.
What the a16z/Mercury report actually measured
The report ranked the top 50 AI-native application companies by observed spending among more than 200,000 Mercury customers. The data came from ACH transactions, IO card purchases and wires during June–August 2025. The methodology and ranking are described in a16z’s report.
“Where startups spend” therefore means which application vendors appeared most prominently in that customer population’s transactions. A high position may reflect broad adoption, large bills, recurring seats, usage charges or Mercury’s particular distribution. The report does not publish a complete dollar-by-dollar market-share table, so rank is not a measure of total industry revenue or product quality.
Recommended Free Tools
#1 Best Overall
What is outside the measurement
- Purchases made through non-Mercury cards, accounts or reimbursement systems.
- Mercury Personal customers and companies that do not bank or transact through Mercury.
- Internal engineering payroll, consultants and other labor costs.
- Cloud, GPU and infrastructure purchases, which the ranking excludes.
- AI embedded inside ordinary software subscriptions when it is not separately visible as a transaction.
- Complete cloud spending: Google Cloud and Gemini were combined because the data could not separate them.
Mercury customers are not a statistically representative sample of all startups. The population may overrepresent venture-backed, technology-oriented, digitally native or U.S.-based companies. Treat the results as a useful purchasing signal, not national accounts.
For independent context, TechCrunch’s coverage likewise describes broad product adoption and a market still tilted toward copilots rather than fully autonomous workers.
The spending map: what startup buyers purchased
| Category | Examples in the ranking | Likely purchase | Important qualification |
|---|---|---|---|
| General assistants | OpenAI (#1), Anthropic (#2), Perplexity (#12), Merlin AI (#30) | Research, drafting, analysis, coding and general reasoning | Transactions do not show frequency of use, productivity or return on investment. |
| Coding and app building | Replit (#3), Cursor (#6), Lovable, Emergent | Prototypes, front ends, feature work and prompt-based product creation | These tools do not establish that engineers can be safely eliminated; testing, security, architecture and maintenance remain separate work. |
| Creative and media | Freepik, Canva, ElevenLabs and other creative products | Images, presentations, campaigns, audio and other marketing assets | Licensing, brand consistency and human quality review still matter. |
| Customer service | Lorikeet (#8), Customer.io (#14), Ada (#40), Crisp (#46) | Ticket routing, answer drafting, search and support automation | A subscription does not prove fully autonomous support; escalation to people may remain central. |
| Sales and go-to-market | Instantly (#13), Clay (#25), 11x (#37) | Prospecting, enrichment, personalization and campaign execution | Deliverability, consent, brand safety and quality controls limit end-to-end automation. |
| Recruiting and HR | Micro1 (#9), Metaview (#19), Applaud (#43) | Sourcing, interview intelligence and recruiting operations | Candidate consent, discrimination, explainability and employment-record controls are material risks. |
| Specialized vertical work | Delve (#11), Crosby Legal (#27), Combinely (#29), Cognition (#34), Serval (#39), Alma (#42) | Compliance, legal, accounting, engineering, IT service desk and immigration workflows | Narrower fit can improve workflow value while increasing integration, accuracy and regulatory exposure. |
The list is therefore a map of work being purchased, not a leaderboard of the “best” models. A startup may use a general assistant for occasional analysis, a coding product for development and a vertical system for support or compliance at the same time.
Horizontal software still dominates the list
a16z classified 60% of listed companies as horizontal applications and 40% as vertical applications. That is a share of companies in the top 50, not a share of spending or revenue. Horizontal products include assistants, workspaces, coding tools and creative software that can serve many departments. Vertical products target a role, industry or tightly defined workflow.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The mix matters for budgeting. Horizontal tools can spread quickly across a team but create duplicate subscriptions and uncertain usage. Vertical systems may be easier to tie to an operational metric, yet they often require more integration, specialized data and contractual review.
Copilots versus “AI employees”
The report divides vertical applications into augmentors, which make people faster or more capable, and substitutes intended to complete workflows end to end. Of 17 vertical companies, a16z says 12 primarily augment humans and five are oriented toward AI-employee-style substitution: Crosby Legal, Cognition, 11x, Serval and Alma.
That 12-to-five split is the clearest corrective to the autonomous-agent narrative. Startups were spending on systems that draft, search, route, recommend and accelerate human work more often than on products that can own an entire job without supervision. The substitution category is emerging, but the observed mix does not show that AI workers have replaced human teams.
Why coding and app-building tools stand out
Replit, Cursor, Lovable and Emergent represent different buying motions. Cursor is associated with AI assistance inside an existing developer workflow; Replit combines building with an integrated hosted environment; Lovable and Emergent emphasize prompt-driven product creation and broader access to app development. Their presence indicates that “vibe coding” moved beyond casual experimentation into startup workplaces.
That is evidence of willingness to pay for faster prototyping and some routine engineering tasks, not proof that production software can be generated safely without engineers. Security review, dependency management, testing, observability, deployment, compliance and long-term ownership remain necessary costs.
Rank #4
What this means for startup budgets and vendor decisions
For founders and operators
- Start with one repeated workflow whose baseline is measurable: cost per resolved ticket, qualified lead, shipped feature or completed review.
- Control overlapping assistant subscriptions. Seat fees can look small while unused licenses and usage-based charges accumulate.
- Set approval, editing and reversal points before allowing an agent to send messages, change records or execute transactions.
- Separate application subscriptions from model APIs, cloud bills and internal labor when calculating the AI budget.
For investors
- Application demand is distributed rather than consolidated around one vendor in every category.
- Workflow ownership and distribution may matter more than access to a generic model.
- High transaction activity demonstrates adoption, not retention, durable margins, product quality or headcount reduction.
For buyers comparing vendors
- Define the job. Specify the task, volume, owner and acceptable error rate.
- Price the whole workflow. Include seats, tokens, minutes, credits, integration work, review time and escalation costs.
- Check human control. Confirm that users can approve, edit, reverse and audit actions.
- Review data terms. Examine training use, retention, encryption, deletion, access controls and contractual limits.
- Test reliability. Measure failures, hallucinations, response quality and escalation paths on representative work.
- Assess lock-in. Ask how data, prompts, workflows and outputs can be exported, and what happens if pricing or models change.
- Verify security. Look for SSO, role-based permissions, administrative controls and relevant compliance documentation.
Choose a general model when the work changes frequently and requires broad reasoning. Choose a vertical application when a repeatable process, integrations and accountability matter more than flexibility. Build internally only when the workflow is strategically differentiating and the company can support security, maintenance and evaluation.
What the report cannot tell you
- It cannot calculate return on investment, usage intensity, renewal or retention.
- It cannot show whether a product reduced headcount or improved productivity.
- It cannot measure AI infrastructure, GPU rental, internal model development or AI features hidden in broader SaaS contracts.
- It cannot establish that a ranked vendor is safe for regulated data or suitable for a particular jurisdiction.
- It cannot be treated as a current 2026 spending survey: the confirmed observation window ended in August 2025.
Nor does the 60/40 horizontal-versus-vertical classification mean that 60% of dollars went to horizontal software. It describes the companies represented on the list.
The practical conclusion
The report does not show that AI hype is empty. It shows where an early durable spending wave is actually landing: general-purpose assistance, faster software creation, media production, support, sales, recruiting and bounded professional workflows. Autonomous “AI employees” appear in the market, but they are still a minority of the vertical companies identified.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →For a startup, the useful question is not which vendor has the highest rank. It is whether a tool lowers the cost or time of a defined job while preserving security, human accountability and workable unit economics.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
