Conviction wants to be your first call.

Sola’s founders spent a summer living two blocks from Golden Gate Park. By the end, according to their investor Conviction, they still could not navigate to it. Jessica Wu and Neil Deshmukh had been indoors, building enterprise automation. It is a wonderfully Silicon Valley detail: proximity to a landmark, almost no contact with it, and a rather different idea of recreation.

The interesting part is what occupied them. Office workers had been promised that software would take over repetitive tasks. Instead, older automation could require specialists and break when an interface changed. Conviction’s account describes a Christmas spent taking shifts in the office to keep a customer deployment running. That is the distance between an impressive demonstration and a business someone depends on.

Conviction has organized itself around that distance. The San Francisco venture firm finances companies built around artificial intelligence, then helps with the practical business of finding people, customers, and computing resources. Its proposition is that understanding models is useful; understanding what prevents a team from shipping is useful too.

02An investor built around one bet

Sarah Guo launched Conviction in 2022 after nearly a decade at Greylock. Her founding essay announced a $100 million first fund and an interest in what she called “Software 3.0”: businesses built by working with foundation models. The scope extended from chips and infrastructure to applications. The organizing principle was a technological shift, with several possible markets growing around it.

In January 2025, Conviction added former Sequoia partner Mike Vernal and announced a $230 million second fund. That figure describes a fund’s capital commitments. It is not the investment firm’s revenue or valuation. For founders, the more immediately useful number is the firm’s published check range: $1 million to $25 million, often as an early or first investor.

The business runs on venture investing: capital goes into young companies in pursuit of eventual investment gains. Guo’s explanation of Embed, its founder program, also names carried interest as her incentive. Founders are the people Conviction serves day to day; the fund’s financial backers are limited partners. Those are different relationships, even when everyone wants the same company to succeed.

03The smaller check has a bigger job

Embed brings that approach down to the moment when a team is barely a team. Its published offer is a $250,000 investment through an uncapped, no-discount MFN SAFE, or at the price of a recent round. A SAFE provides rights to future equity; the check should not be mistaken for free money. The page also advertises cloud and other infrastructure credits.

The intended class has 10 to 12 companies. Solo founders and teams that have already raised can apply. A San Francisco gathering is required, while other community events are optional. These particulars matter: a program’s usefulness depends on whether its actual format fits the people trying to build.

The larger idea appears in Guo’s essay explaining Embed. A firm can make only so many concentrated investments, yet introduce a wider group of founders to recruits, prospective customers, and experienced operators. It can also share access to dedicated H100 GPUs at cost. The essay describes that arrangement historically; it is not a current hardware price list.

Weekly dinners add another ingredient: accountability. A founder who has to describe progress to other founders has a reason to make some. My reading is that Embed packages a set of small advantages whose value rises when they arrive together. A useful introduction lands differently when a team already has the resources to act on it.

04Follow the work, then the software

The portfolio makes this argument more concrete. Baseten provides infrastructure for running AI models in production. Mistral develops models and associated AI products. Sierra builds customer-facing agents and uses outcome-based pricing. They sit at different levels of the same market, connecting technical capability to work a buyer needs completed.

Then there are the less glamorous bottlenecks. Conviction’s investment note on Latent describes administrative obstacles between a diagnosis and access to medication. Its Corridor note identifies a security problem: software development accelerates, while reviews and vulnerability fixes struggle to keep up. These are specific places to look for a company, rather than broad instructions to sprinkle AI over an industry.

Harvey supplies a dated signal of market appetite. In March 2026, the legal AI company announced $200 million of financing at an $11 billion valuation, with Conviction participating. That is evidence of continued investor demand for one portfolio business. It does not tell us what Conviction has returned in cash to its own investors.

This is where specialization earns its keep. A founder can seek capital from a generalist venture firm or apply to a broader accelerator such as Y Combinator. Conviction’s distinction is the concentration of its attention on AI and the supporting network it builds around that choice. Whether that makes it the right investor depends on what a particular team needs.

“We have the courage of our own convictions.”

Conviction’s public declaration

05The part you can borrow

Conviction makes some of its thinking accessible without an investment. Its “Plausible Schemes” page publishes startup ideas and the obstacles behind them. Mixture of Experts brings candidates together with startup founders for demos, pizza, and conversation. The jobs page offers another route into portfolio companies. Guo also co-hosts No Priors with Elad Gil, interviewing builders and researchers.

The ideas page is especially revealing because its enthusiasm comes with friction. In laboratory robotics, for example, Conviction points to long validation cycles and buyers who fear spoiled experiments. A machine that moves faster will struggle if the person buying it cannot trust what happens next. Technical progress changes the available tools; customer incentives still shape adoption.

A reader can copy the discipline behind that observation. Choose a task, identify the person responsible for it, and find the first point where a demonstration becomes unreliable or uneconomic. Measure the work completed, including the cost of inference and human intervention. Then ask which scarce resource would speed the next experiment: a hire, a customer conversation, compute, or advice.

Credits eventually run out. Introductions need follow-through. A friendly room cannot supply demand for an unwanted product. Conviction’s useful wager is that a young company can make better use of its time when those practical constraints get attention early. Sola’s founders may have missed the park. The customer deployment still needed someone awake.