Creator perspective · MoneyAn AI agent's recommendation carries the full context of the application it runs in, whereas a generic chat LLM only knows a small token amount from memory and current chat — so agent recommendations are far more opinionated and specific. When an AI agent gives a definitive single-vendor recommendation, non-expert users will adopt the top pick instantly without testing alternatives. Generic ChatGPT-style searches tend to recommend the biggest brand (e.g., Vercel), while in-app agents recommend the best-fit tool — so agent recommendations can surface lesser-known vendors. Winning how AI agents pick applications matters more than winning how humans pick via LLM chat (AEO/GEO).
None; argued from the speakers' experience.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 882: Claudeforce vs. Replit vs. Lovable. Who wins? (The Agents #016)
Creator perspective · MoneyThe new go-to-market playbook is 'ask your agent' rather than 'ask your developer' — agents now make tool-selection decisions. Jason's Connect agent chose a hosting company he'd never heard of, and Amelia's 10K agent told her to build AEO tooling rather than buy it.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 882: Claudeforce vs. Replit vs. Lovable. Who wins? (The Agents #016)
Principle · MoneyThe host argues you've got to win the war of agent recommendations — not just the battle of ChatGPTs. They coin agent-led growth (ALG) as the term for agents choosing products, and say agent-first marketing should be called AFM, not AEO or GEO. It matters more for the 10K agent to see your app than maybe even the CEO. If you think agents asking about your product doesn't make sense today, it will kill you next year. Salesforce's marketing team needs to step up on agent-first marketing.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 882: Claudeforce vs. Replit vs. Lovable. Who wins? (The Agents #016)
Creator perspective · MoneyAgents you connect can act on their own with real risk, so guardrails are needed: install a secret-keyword 'safe word' before an agent can take actions on another system, verify a data platform keeps daily snapshot backups outside checkpoints, and be aware agents may create hidden communication channels (shared inboxes, REST endpoints) you don't fully control — check whether they exposed them to you voluntarily.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 882: Claudeforce vs. Replit vs. Lovable. Who wins? (The Agents #016)
Creator perspective · MoneyWhen you ask Claude in general what app to pick for CRM, it only has a little context from the chat itself, a small token amount in memory, maybe searches other chats, and maybe something deeper in a project — that's all it knows when it makes a recommendation. When you ask an agent inside an application, it has all the context of whatever it's in, so it is very opinionated. A generic ChatGPT search would probably tell you to use Vercel because it's the biggest brand.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 882: Claudeforce vs. Replit vs. Lovable. Who wins? (The Agents #016)
ReportThe host reports that a 2024 team from University College London and UC Davis wired a Samsung TV and an LG TV and monitored network traffic while watching regular content. ACR data was sent in every case tested — antenna, HDMI laptop, game console — except screen mirroring from a phone, sitting idle on the home screen, and using Netflix or YouTube smart apps.
Slidebean · The Most Profitable Part of Your TV
Principle · MoneyThe founders put their kids first and built the business around fatherhood, doing it for a deeper purpose rather than just money. They argue that rapid hypergrowth requires so much work that it takes you away from your kids' key moments, so pouring gasoline on hypergrowth can be counterproductive when it pulls you from what matters most. They believe having time away from the business to be a present father and husband will pay off in the long term — calling it a thing that doesn't scale — and note that working from home and building online now lets them be with their kids all the time.
Foundr · We Started Dad Gang with $750 and One Hat - and did $40M in 4 Years
Principle · MoneyA big part of community building is giving people an identity and having them wear the product proudly.
Foundr · We Started Dad Gang with $750 and One Hat - and did $40M in 4 Years
Principle · MoneyThe founders chose to be husbands and dads first; aggressive growth can be counterproductive if it takes you away from your kids, who are the most important thing.
Foundr · We Started Dad Gang with $750 and One Hat - and did $40M in 4 Years
Principle · MoneyThe founders started Dad Gang while working full-time jobs, with no business plan, no pressure to grow quickly, and an idea that felt organic and authentic to who they are; they recommend identifying the common denominator across your past ventures and interests (for them, fatherhood) to find a business that fits naturally, and creating products you personally love to wear or use.
Foundr · We Started Dad Gang with $750 and One Hat - and did $40M in 4 Years
Principle · MoneyExhaust contract work as much as possible before hiring full-time staff; businesses that hire too fast get into trouble when numbers don't keep up.
Foundr · We Started Dad Gang with $750 and One Hat - and did $40M in 4 Years
Principle · MoneyThe speaker observes that even with AI, the vast majority of people will still be dealing with people — teams and other humans — and it's the messiness of the emotions that makes dealing with people so hard. You already know logically what you need to do about an underperformer; the delay between knowing and doing is what delays the success, growth, team, and culture you want. Shrinking that delay is what makes you a better entrepreneur, and the best entrepreneurs he has seen act almost immediately — the moment they know something, they do it. He notes this is why some of the highest-level entrepreneurs get a reputation for being sociopaths or not caring about people, and why people tell stories of individual firings (like Steve Jobs, Zuck, or Elon firing someone ruthlessly) — a story of a
Alex Hormozi · My honest advice on how to raise your team’s bar
Principle · MoneyThe speaker says the single direction 100% correlated with a company's growth is the entrepreneur's tolerance for mediocrity and the quality of the team they surround themselves with. He adds that as you win more, what was once good becomes mediocre — people need to grow with you or get out — and that nothing turns off high-quality people like low-quality people.
Alex Hormozi · My honest advice on how to raise your team’s bar
Principle · MoneyThe speaker distinguishes true losers, who leave on their own because they are so painful to keep (decent owners get them out anyway), from the mediocre middle: people who are slow and produce mediocre work that winners have to rework. He identifies clock-in, clock-out behavior as what creates mediocre organizations, noting that the vast majority of organizations run this way. His stance: let people live how they want, just not in your business.
Alex Hormozi · My honest advice on how to raise your team’s bar
Principle · MoneyThe guest argues that if AI is adopted and utilized properly, it will make a firm more competitive. In actively traded markets, competitiveness comes down to your process and the information at your fingertips — knowing you have a strong process to execute and knowing your available liquidity and pricing. In private markets, better information access is evolving and will turn into a competitive advantage over time, and firms with access to better, accurate information and teams that know it is accurate will get further ahead.
AI Engineer · AI Agents Cut 3-Day Portfolio Builds to Minutes — Jack McDonald, Apollo & Michael Manapat, Rowspace
Principle · Moneyswyx states that a core skill of the AI engineer is defining which loops you are working in and figuring out whether that is the right level of abstraction.
:0.1
AI Engineer · The Highest Loop — swyx
Creator perspectiveBecause chemists have distinctive chemical styles (trainees partially adopt mentors' styles, and chemists' unrecorded decisions not to synthesize act as fingerprints), models trained on historical data may learn 'the sociology of the problem' rather than ground-truth chemistry. Even self-generated data is likely confounded, since organizations use different chemists for different project types. Superstitious preference for building blocks from prior clinical programs is a specific anti-pattern models learn as a proxy for efficacy.
Latent.Space · Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio
Creator perspectiveDNA-encoded libraries are the most scalable technology for measuring small molecule interactions because millions of molecules can be pooled in a single tube: attach a DNA barcode to each small molecule, pool huge numbers, dunk the protein in like a fishhook, pull out binders, and sequence the DNA barcodes. When coupling life science with ML, there are really only two scalable readouts — pictures or DNA sequencing (sequencing is just another way of taking pictures). Sequencing is dirt cheap, extremely scalable, and getting cheaper again after a plateau, with implications for data collection scale.
Latent.Space · Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio
PrincipleLeash Bio is extremely bitter-lesson-pilled, believing in scaling and getting signal from scalable measurement regimes; if you can get pretty good signal out of a scalable measurement regime, you should just go get that, and if it is hard and you have specialized knowledge, that is the time to pour gas on it.
Latent.Space · Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio
Creator perspectiveShri Kusuri's tweet poses the machine-learning-for-drug-discovery business model paradox: building a good ML model requires lots of high-quality data at scale, but if you can already collect that data well, what do you need the model for? Quigley's response: if your only goal is good chemical material for a particular target, you don't need a machine learning model — just collect the data. The model only adds value if you have broad chemical material across many targets to drive deeper chemical understanding and faster optimization.
paradox quoted from a third party; resolution is Quigley's interpretation
Latent.Space · Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio