Home›Companies›General Agency

AI coworkers that can learn and act just like a human

We are building Tessa, an AI coworker that can autonomously learn and master the usage of web-based tools, just like a human does. To accomplish this, we are building a novel neural architecture that enables autonomous self-improvement in AI agents. So far, we've achieved a state-of-the-art success rate of 93% on the WebVoyager benchmark. We are working to automate workflows in several verticals, including property management & real estate, financial services, and consulting.
Active Founders
Mo Nasir
Mo Nasir
Founder & CEO
Co-founder and CEO of Tessa, an AI coworker that can learn and master new skills like a human does. Previously worked on self-driving cars at NVIDIA. Chinese loose leaf tea connoisseur 🍵
Harvey Hu
Harvey Hu
Founder
Co-founder and CTO of General Agency. Previously machine learning engineer at Google & TikTok, designing recommendation algorithms for billions of users. I love models, probabilities, and solving complex impactful problems
Company Launches
Tessa API: State-of-the-Art Browser Agent for Your Vertical AI
See original launch post

If you have tried building browser agents into your vertical AI application, you know how tricky they are. The last thing you want is for the agent to “act up” with your browser and do something other than what you asked for. You need web agents that’s smart, reliable, and accurate.

You need the best one out there and that agent is Tessa, according to the latest benchmarks.

State-of-the-Art Results

For 8 months running, we have maintained state-of-the-art performance among browser agents. We first achieved SOTA results of 93% success rate on the Web Voyager benchmark in December 2024, before OpenAI's Computer Use Agent and the current wave of browser agents entered the market.

Now, with intense competition and industry attention, our latest benchmarking shows we are still leading by a significant margin.

The REAL benchmark evaluates web agents in complex, real-world scenarios and is scored on mostly classical, deterministic criteria. The latest results show Tessa achieves 54.5% success rate—a 12% lead over Anthropic’s computer use. If you are looking for a browser agent for your vertical AI application, Tessa is your best bet to get the task done.

What we offer

  • Simple to get started: just give it a directive
  • Bring Your Own Browser (BYOB) or have Tessa spin it up for you
  • Stealth built-in. The agent’s actions are indistinguishable from humans
  • Choose the best LLM models

We benchmarked different models’ performance against the cost in the following two graphs for you. We recommend sonnet-4 for most cases but Gemini 2.5 Flash, o3, and Opus 4.1 are all good choices along the price-performance boundary.

How did we beat OpenAI and Anthropic?

We pioneered a unique inference-time scaling architecture for browser agents, named RAVE (Reason, Act, Verify, and Evaluate), which helped us maintain the lead. Read our technical deep-dive blog post for the full methodology here.

Start Building with Tessa Today

We have been using it to build transformative workflows for our clients over the past few months. Now you have access to it too through a simple API.

url = "https://api.heytessa.ai/v1/run_browser_agent"
headers = {"Authorization": f"Bearer {YOUR_API_KEY}"}

result = requests.post(url, headers=headers, json={"directive": "Go to news.ycombinator.com and get the titles and points of the top 3 stories on the homepage"}).json()
print(result['history_url'])

We Support Any Vertical

Legal AI: Research case law across Westlaw, LexisNexis, court databases, or your internal databases to retrieve information.

Financial Applications: Aggregate real-time data from trading platforms, regulatory filings, and financial databases.

Healthcare: Navigate patient portals, insurance systems, and medical databases with the reliability healthcare workflows demand.

E-commerce Intelligence: Monitor competitor pricing, inventory, and product data across hundreds of sites.

Research Platforms: Gather information from academic databases, journals, and specialized repositories.

Try our API today and see why choose Tessa for production web automation.

Get API Access → | Book a Demo →

Previous Launches
The easiest way to automate work using natural language
Self-improving AI agents to automate backoffice business processes
General Agency
Founded:2024
Batch:Winter 2025
Team Size:2
Status:
Active
Location:San Francisco
Primary Partner:Aaron Epstein