Randy Hunt recommends that working examples of good design speak better than trying to shape perception through communication, pitches, or compelling stories — he would rather shape perception through the visceral reaction to something being very good. Successful case studies and working partnerships are the vehicles that change how organizations perceive design.
Sujee explains you need an engine to serve the model, that not all models run equally well on all engines, so you should figure out which engines run well and deploy models accordingly — on the best engine possible rather than a default engine. Nebius utilizes a lot of open source engines and maintains their own internal fork that they continuously tweak and optimize.
Carmen Li explains that Compute Exchange by design does not own GPUs, which avoids conflicts of interest with clients and providers. If a marketplace owned GPUs, it could cherry-pick clients and hold relationships, like a NASDAQ that only traded its own stocks.
This is the creator's own design rationale for her company.
Carmen Li recounts that the secondary server transaction market started last year when people with excess inventory (e.g., 20,000 H100s) asked her to take them off their hands. The market has been vibrant for the past 12 months, covering all servers including B200, with secondary transactions sometimes trading above MSRP. She argues this can make sense, analogous to COVID times when used cars traded above new car prices. The market covers 409Ls, DB200, AMD, and everything, with live inventory and connection through channels like Slack, email, and chat.
Market observations from the speaker; some figures carry possible numeric error flags.
American restaurant owners receive emails from wholesalers, probably once a month or once a quarter, with an Excel spreadsheet or PDF of prices for thousands of inventory items. With an AI agent, a restaurant owner can prompt their way to having a wholesaler price-change analysis exist for them in five minutes. The owner doesn't care how often the agent runs — what they care about is occasionally getting an email telling them to move purchases to another wholesaler or drop menu items that got too expensive.
hypothetical/illustrative example
When thinking about AI agents, people focus too much on reducing friction and toil and making things as efficient as possible, but humans don't always optimize for efficiency. Some friction is enjoyable to humans as an experience, while other friction is not. On Instagram, users don't want efficiency — they want to be entertained. There are many times when users are optimizing for friction and do not want ads or marketing in that experience.
speaker's conceptual argument
Building a product designed for the user requires building it for the user, not for corporate overlords and lawyers, and eventually working backwards into enterprise sales and sales-led growth.
endorsed by the speaker as an argument
Randy Hunt argues that push notifications are as much of a design opportunity as a core dashboard screen; that social discovery, service emails, and push notifications are both product and brand touchpoints; and that creating all of them is the full canvas — which makes the work fun, like a playground.
confident
Randy Hunt says it is extremely helpful to allow people to have a space to initiate, drive, push through, get out the work, and enjoy the responsibility of making it successful — and that maintaining conditions of trust and ownership over time is extremely satisfying to people.
confident
Randy Hunt argues that talent density reinforces the other mechanics of small teams and that you have to get the flywheel going: people with a similar mindset see the talent-dense team and want to join, so hiring also compounds in that direction.
Randy Hunt, after trying many team configurations, prefers small teams primarily because of lower communication overhead: every added person brings new interests, motivations, and conflicts, while small teams make it easier to hold shared understanding and spend time on the work itself rather than the work around the work.
Randy Hunt describes a shift in some organizations where designers are becoming more technical and shipping more code in the production application, a change underway even before coding agents existed.
hedged
From Milton Glaser's dismissiveness toward his student work — 'Well, that's not very good' — Randy Hunt took the lesson to keep the floor of quality really high so the ceiling can keep rising, a lesson he says he has appreciated across many chapters of his professional life. He also interprets Glaser's behavior as giving time and attention only to work that meets the top standard, 'teach to the top of the class.'
Randy Hunt recommends that building a talent-dense small team starts with clarity of intent: being explicit that the team will be smaller than historical or outside reference points, that this is intentional, and that it works.
Tenex built its own CRM and its own OS from scratch, and learned through that that maintenance overhead is a real cost of self-built software. Alex Lieberman ran a 'Kill My SaaS' competition replacing a mid-tier SaaS: quoted $40,000/year, replaced for $1,000, going to production for the next conference — and it's going okay so far. When replacing a mediocre SaaS product, ask what the 10x better version would look like — as the customer, you can iterate in minutes, not months. Evaluate both whether you're being overcharged AND whether the product has been mediocre for a long time. Startups and high-growth mid-sized companies can be nimble enough to build 10x better versions of the SaaS they use; enterprises entrenched in a capability rarely will.
The biggest challenge with output-based pay is compensating engineers for time not spent coding (architecting, client meetings). Tenex has had to find nuance in the comp model, but it is still largely driven based on output.
A 10x smarter model rarely justifies its cost for low-value tasks like moving cells in a spreadsheet — most economically valuable tasks can be done with Opus 5.5. CloudTag and similar tools provide value mainly for teams collaborating on non-technical tasks; don't expect them to replicate a developer's local agentic workflow.
As a new-entrant founder, your advantage is speed — build a different product and use different go-to-market approaches before incumbents copy you.
endorsed
Don't enter a market where incumbents can use the new technology to serve their existing customers faster than you can grab market share — you'll lose the race, as the guest realized with tax incumbents.
discouraged
For external-facing roles, email and calendar are real sources of insight about how to automate part of someone's work.
described
No published protocol matches this selection.
The creators behind most of what cleared.
195 cleared insights match
99 cleared insights match
85 cleared insights match
Choose sources → to get a library of your own and a daily brief.
