Selling back overbought compute
Carmen Li advises that if you exceed your needs in quantity or duration, you can sell the excess compute back to Compute Exchange.
Results
WHY · 421 of 33209 results · nasdaq
Frame agent adoption around reallocation and elevate SMEs above the work
When introducing AI agents to customers, frame the conversation around reallocation: what can we replace, and what are the next biggest problems for those people to work on? In most large companies, the cycle of replacing automated work and finding the next biggest problem continues forever. When AI agents take over routine work, bring deep subject-matter experts 'above water' to work across teams and think about how parts of the business work together, rather than selling in a spreadsheet. Elevate finance-division employees who are mostly in spreadsheets to think strategically about the business and where money is allocated.
Run technical due diligence before building AI automation
Run technical due diligence to prove an AI automation is feasible before committing to build it — the guest describes doing this as an engineer before building their tax automation.
described
Frame AI positioning around concrete dollar outcomes
Frame AI product positioning around concrete dollar outcomes ('save $500 a year, charge $5') rather than capabilities, because the answer to a concrete savings offer is an easy yes.
endorsed
Chase user pull for a few months before asking about TAM
When validating a startup idea, chase user pull for a few months before asking how big the TAM is or what winning means.
anecdotal
Look for simple common knowledge-worker workflows no one has built AI around
Look for common knowledge-worker workflows that are simple but that no one has built AI around — meeting prep, meeting outcomes, answering simple repetitive emails, and day preparation.
described
Mine existing customers for repeated pain points when searching for new ideas
When searching for a new startup idea, mine existing customers for repeated pain points — tax customers' complaints about meeting prep and repetitive email answers revealed the agent opportunity.
described
Only ask for deeper system access after demonstrating value
Only ask for deeper system access (CRM, IT tools) after you've demonstrated enough value that the user feels the magic.
endorsed
Pick AI startup markets with lots of text
When picking a market for an AI startup, look for domains with lots of text, because LLMs are very good at text — the guest cites tax as an example.
described
When you find product-market fit, go hard and keep pulling the string
When you find product-market fit, hang on for dear life and keep pulling the string — don't pause to ask how big the company could be or what winning means; go hard for a few months and learn about the market.
endorsed
Provide audit logs, privacy posture, and cost controls when selling AI agents to enterprises
Provide audit logs, privacy posture preventing privilege information sharing with agents, and cost controls for the CFO when selling AI agents to enterprises.
endorsed
Onboard AI agents with email and calendar access only first
Onboard AI agents by starting with email and calendar access only — not core systems of record — to demonstrate value before requesting deeper integrations.
endorsed
Sell automation by the outcome, not the job title of the tool
Sell automation by the outcome the customer wants (e.g., 'save money on iceberg lettuce', 'save 15%'), not by the job title of the tool (e.g., 'do you want an executive assistant?') — customers don't recognize themselves in the abstract role.
endorsed
Compute Exchange as a marketplace for buying GPU compute
Carmen Li describes Compute Exchange as a marketplace where buyers can browse verified compute resources and submit requests, analogous to using NASDAQ to buy stocks.
Tenex is an applied AI partner helping large companies go AI-native
Tenex is an applied AI partner for some of the biggest companies in the world, helping them move from using AI apps to becoming AI-native. It started 18 months ago as an AI-powered dev shop and is now at 75 people, with 90% of employees being engineers or AI strategists (technical consultants).
The biggest diffusion gap: agents in production for non-engineering work
The biggest diffusion gap between raw AI intelligence and enterprise usage is agents in production for non-engineering work — most enterprises with 1,000+ employees have no agent in production doing non-software work. One of the most common enterprise asks is help building an agent for a back-office task, which is why Tenex now deploys agents into production outside the engineering org for companies' repeatable work.
Thesis: CIOs/CEOs will want consumer-grade AI assistant experiences; businesses weigh convenience and enterprise features together
The speaker's thesis is that CIOs and CEOs will want the consumer-grade AI assistant experience for their company, and whoever has an enterprise experience that feels like a consumer experience will be set up to win. Businesses will care about both the convenience and the enterprise features when choosing an AI assistant product. The company wants to be positioned as the business with an enterprise experience that feels like a consumer experience, and has mostly succeeded in building that; the next question is how to go from one enterprise deal to dozens effectively.
AI-for-business is a bigger market than coding, and will move the needle within 12 months
The market for AI doing more for a business and helping the business be more successful is bigger than the coding market. The speaker believes their horizontal market is actually the biggest thing longer term, even though short term it is not as urgent as coding. AI helping businesses will not move the needle at companies in the current environment but will within 12 months.
speaker's belief; the 12-month timeline is a prediction, not a fact
As models improve, FDE gets worse while automation scales better
As AI models improve over time, the FDE approach actually gets much worse, while automation-based approaches that use AI to build custom software scale much better.
speaker's argument/heuristic, described with 0.85 confidence
Startups must build for a less-human-in-the-loop future
When building a startup, you always have to think about what the future looks like, and the speaker believes the future looks like a lot less human in the loop than today. The speaker's job is to get companies comfortable with a less-human-in-the-loop universe because the economic benefit far outstrips any risks they would see in that experience today.
speaker's belief and argument
Protocols you can run
WHY · 0 of 59 published protocols
No published protocol matches this selection.
Source analyses
WHY · 5 of 627 current source analyses
People worth following for this
The creators behind most of what cleared.
WHY · 3 creators you don't follow account for 379 cleared insights
- The a16z Show
195 cleared insights match
- Design Better
99 cleared insights match
- Latent.Space
85 cleared insights match
Choose sources → to get a library of your own and a daily brief.