On Thursday, Andreessen Horowitz launched its first AI Spending Report in partnership with the fintech agency Mercury. Utilizing transaction information from Mercury, the report analyzes the highest 50 AI-native software layer corporations that startups are spending cash on, just like the beforehand revealed Prime 100 Gen AI Client Apps. 

A16z companions Olivia Mooe and Seema Amble say the information exhibits corporations are nonetheless adopting a variety of various AI merchandise for sure duties — and new apps are rising and falling in a short time.

“There’s a proliferation of instruments,” Amble mentioned. “It hasn’t simply coalesced round one or two in every class.”

The report additionally exhibits numerous spending on “human augmentors” or “copilots” that may assist increase productiveness among the many workforce, suggesting startups aren’t prepared to completely shift into agentic workflows.

“As pc use turns into extra of a mode and there’s extra of the power for there to be end-to-end agentic flows constructed, I feel that shift will occur, the place we’ll see fewer copilots and extra end-to-end agent instruments,” Amble mentioned. “Particularly as individuals are actually keen to provide them a strive.” 

Unsurprisingly, the highest of the record was dominated by main labs, with OpenAI taking the highest slot and Anthropic following up at quantity 2. Vibe-coding instruments had been additionally effectively represented, with Replit at quantity 3, Lovable at quantity 18. Cursor landed at quantity 6, and Emergent at quantity 48. Cognition, which operates extra enterprise-oriented coding instruments like Devin and Windsurf, was at quantity 34.

When a16z produced an analogous record for client habits, Lovable ranked a lot increased than Replit on pure site visitors alone as a result of lots of people had been utilizing it to create tasks. However startups are usually not spending as a lot cash on Lovable as they’re on Replit, partly due to the dearth of enterprise options. However the number of corporations on the record appears to recommend there’s room for loads of completely different corporations without delay.

“It’s an open query going ahead on vibe coding,” Moore mentioned. “Does the area begin to consolidate, and one place turns into one of the best platform to vibe code? Or is it the case the place there’ll be 4 or 5 extra vibe coding corporations which are actually massive companies for several types of purposes? We don’t have the reply to that but.” 

Moore was additionally shocked to see startups adopting consumer-oriented instruments like CapCut and Midjourney. 

“We’re seeing that numerous these [consumer] corporations are getting yanked into enterprise quicker and quicker as a result of they make such pleasant client instruments that then individuals undertake and use as people and convey into their groups and workplaces,” Moore mentioned. 

Horizontal purposes general made up not less than 60% of the names on the record, and 40% consisted of vertical purposes. The most well-liked vertical software program corporations fell into three buckets: gross sales, recruiting, and customer support. However the report additionally discovered AI making progress in lots of sectors that earlier generations of startups had struggled to crack.

“What possibly beforehand would have been, like, service corporations or consultancies are actually software program corporations within the age of AI,” Moore mentioned.

Amble gave Crosby Authorized for example, which may rapidly assessment a authorized contract for a consumer, changing what at one level would have been a gathering with an in-house normal counsel, rummaging by way of ideas and analysis. She mentioned proper now, a lot of the instruments are used to assist staff (like a co-pilot) in making choices quicker, moderately than changing total workforces and expertise fits with automated employees (end-to-end brokers). 

“Because the tech will get higher, and we’re truly capable of construct out full agent co-workers, you’ll see that blend shift extra towards end-to-end brokers and away from co-pilots,” she mentioned, later including that AI instruments can do way more work, like outreach, quicker than a human can. 

There have been additionally numerous word takers on the record, reminiscent of Otter.ai, Retell AI, and Habbyscribe — with no single possibility dominating. That is what Amble meant when she mentioned there isn’t one product but that has dominated the market; moderately, startups are nonetheless selecting “their very own taste” to see what instruments they like finest. That is additionally good for workers, who, with so many choices, can decide what purposes assist them work finest, moderately than utilizing a one-size-fits-all product “pushed down from the highest,” Amble mentioned. 

The final massive discover within the report was the growing intertwining of client and enterprise companies. Individuals are bringing the non-public purposes they use at dwelling to work, and individuals who have began corporations are utilizing their favourite private purposes to assist construct their companies. Earlier than, there would have been a delineation between the 2: a set stack for what to make use of whereas constructing a startup. 

Amble and Moore cited Canva for example: It’s a preferred client app that additionally has a large enterprise viewers. It took years for Canva to even add an enterprise plan. However as particular person and enterprise use circumstances change into tougher to differentiate, corporations are extra keen to mix the 2. 

“Your TAM [total addressable market] is not one or the opposite, however you possibly can promote into each,” she continued. She mentioned corporations constructing these merchandise may also “professionalize” quicker, that means constructing out enterprise groups, like go-to-market, gross sales, and help, to allow them to begin promoting and accruing enterprise income sooner, moderately than relying on particular person shoppers. 

Moore and Amble expect the record to alter rapidly within the coming years. Older corporations are actually launching AI options to remain related and accessible, whereas new entrants arrive with new concepts.

“Legacy gamers, legacy nearly means, ‘what was 12 months in the past,” Amble mentioned. “If we pull this once more in 12 months, will the identical note-taking apps even be on there? Or will there be an entire new set?”