Adsmith is a clear example of an agentic workflow applied to a high-value business vertical: digital advertising. While many AI tools focus purely on generation (creating the image) or analysis (reading the report), Adsmith functions as an autonomous manager. It closes the loop by generating creative, deploying it to live platforms via API, measuring the result, and rebalancing the budget based on statistical probability. It is active in the 'execution' layer of the agent stack, where the agent has permission to spend real-world capital and make operational decisions without human intervention for each action.
For the broader AI agent ecosystem, Adsmith demonstrates how specialized agents can dismantle the 'service agency' model. It moves the technology beyond a copilot for marketers and toward an autonomous utility. Their focus on 'closed-loop learning' is a blueprint for how agents in other fields—like sales or supply chain management—might eventually handle complex, data-heavy tasks that require both creative output and financial optimization.
Adsmith (formerly known as Lemnis) is built on the premise that performance marketing has shifted from a creative problem to a statistical one. In the traditional agency model, a team of humans drafts a handful of concepts, tests them over several weeks, and manually adjusts budgets based on intuition or basic A/B tests. Adsmith replaces this entire workflow with an autonomous system that automates over 200 creative decisions and runs experiments at a scale humans cannot manage. The company claims to deliver twice the performance of a human agency at half the cost by focusing on the rate of learning rather than the quality of a single 'hero' creative.
The core of the platform is a closed-loop system that uses performance data from prior experiments to improve future predictions. Instead of launching one broad campaign, Adsmith breaks audiences down into what they call "nano niches." The system generates personalized assets—including images, captions, and keywords—tailored to specific interests, locations, and languages. For example, a customer using Adsmith to launch in five different countries can auto-localize creative without speaking the native languages, as the system handles the translation and cultural context through its generative engine. This level of granularity allows brands to capture the collective attention of many small niches that would be too expensive to target manually.
Founded by Misha Leybovich, the team behind Adsmith includes engineers and data scientists with backgrounds from SpaceX, Google, MIT, and Berkeley. This engineering-first culture is reflected in their "theses": human decision speed cannot keep up with modern ad platforms like Meta Advantage+ or Google Performance Max. They view marketing as an always-on utility that should be constantly experimenting. This mindset moves the focus away from "winning" ads and toward a winning process of discovery. By automating the "kill losers and boost winners" cycle across channels, the platform seeks to eliminate the waste inherent in traditional campaign management.
Adsmith has structured its business model to directly challenge the agency retainer. Most agencies charge a flat monthly fee or a high percentage of ad spend (often 15-20%) regardless of performance. Adsmith operates on a self-serve basis with no retainers, no commitments, and no minimum spend. They charge 10% of campaign spend plus a nominal fee of $0.10 per image generated. For companies that want a completely hands-off experience, they offer a managed service for $1,000 per month. This pricing model aligns their incentives with the advertiser; they only make more money if the advertiser is successful enough to increase their spend. Early customers like Equip Foods and Brilliant Labs report lower CPAs and faster creative iteration, suggesting that the platform is finding traction with growth teams that have outgrown manual experimentation.
An AI-native marketing system that automates creative generation, testing, and budget rebalancing.