Creator perspective · MoneySeveral demand-side attacks: the foundational LLMs themselves may absorb your product and turn you into an obsolete wrapper; ask whether your customer will still exist at all (e.g., translation/transcription agencies whose revenue is effectively going to zero); if you charge per seat and AI handles 80-90% of the work, you keep only 2-3 monitoring reps instead of 6-15, losing 70-80% of revenue; if the people who use your product (e.g., receptionists) are going away entirely, your market goes to zero; and ask whether the step you manage or the artifact you create still matters. If demand goes to zero or gets cut in half, that's a big shock.
Startups For the Rest of Us · Episode 855 | The Eight SaaS Moats that Still Work in the Age of AI
Principle · MoneyOnce Anthropic and OpenAI show you the type of model you need to build, it is a lot easier to figure out what to do — fast following becomes very doable.
described, 0.85 confidence
The a16z Show · Can Open Models Keep Up With the AI Frontier? | Nathan Lambert
Creator perspective · MoneyTo handle conflicting bank package-version requirements, Al Chen endorses creating a skill that does the heavy lifting, reporting, and internal research. Its output is a table format that tells the engineering team which CVEs must be fixed and provides explanations to bank security teams about what cannot be fixed due to transient dependencies.
endorsed
AI Engineer · Shipping AI Into 5 Major Banks Without a 12-Month Compliance Wait — Al Chen, Cisco
Report · MoneySam describes that reaching out to a profile viewer at a company with an active opportunity is a fantastic way to multi-thread into the deal, and that it may lead to accelerating the opportunity or an expansion.
30 Minutes to President's Club | No-Nonsense Sales · #614: How to Use LinkedIn Sales Navigator for Lead Generation
ReportPamela Fox demonstrates testing image input by sending local files to LLMs and asking questions about the files, using an image of an aurochs (not a unicorn), a photo of crocodiles in Costa Rica for distinguishing alligators from crocodiles, and a screenshot of a table with small fonts listing plants for reading small text. She asks each LLM to interpret each image to test multimodal understanding, noting DeepSeek only accepts text and is not a multimodal model, so it is excluded from image input tests.
AI Engineer · The Model Swap Workshop: GPT-5.5 vs Kimi vs DeepSeek vs Mistral — Pamela Fox, Microsoft
Creator perspectivePamela Fox notes it is really hard to bootstrap evaluation before going into production, yet evals are needed before production; she suggests A-Cert helps bootstrap evals and build a test set pre-production. She also observes people keep being told to run evals but no one runs them, or they run only default evals.
AI Engineer · The Model Swap Workshop: GPT-5.5 vs Kimi vs DeepSeek vs Mistral — Pamela Fox, Microsoft
Report · MoneySam McKenna describes how the best reps parallel process vendor review during the evaluation, so that by the time the decision happens, roughly 95% of the paperwork is already done.
30 Minutes to President's Club | No-Nonsense Sales · #613: Master Parallel Selling Strategies to Speed Up Every Deal | Sam Mckenna | 30MPC Hall of Fame
PrincipleThe hosts describe how Seinfeld's creators got a sitcom pilot approved by routing it through NBC's late night and specials department instead of the comedy department — borrowing the familiar sitcom format to accomplish a novel goal. Jerry quoted: 'every step of the way, we were waiting for somebody to come in and say, you can't do this on network TV. But nobody did' — when pitching something genuinely different, expect no one to object precisely because it doesn't fit existing categories.
Acquired · Seinfeld
ReportLarry and Jerry wrote The Contest in secret — not on their shared office whiteboard where other writers and NBC executives could see plot points — keeping the script secret throughout the writing process until the table read, which was as close to the point of no return as they could engineer before revealing what they were doing.
described
Acquired · Seinfeld
ReportAfter The Contest, NBC decided the show was magic and special and bet the network's future on it, moving Seinfeld to Thursday immediately to air after Cheers, with the plan that when Cheers went off the air Seinfeld would take the top slot if it performed.
described
Acquired · Seinfeld
Creator perspectiveWhen a creator blows up at a network executive and threatens to quit over their vision, that behavior is so rare that nobody did it — no creator before Larry David would threaten to quit when the network objected — yet it worked for Larry. Typically, creators cave and do whatever the network wants; Rick Ludwin's softer approach was walking Larry around the lot to explain concerns rather than issuing a formal note.
described
Acquired · Seinfeld
Report · MoneyA rapid succession of agentic payment standards has launched: L402 by Lightning Labs (the first, very early), Mastercard AgentPay (April last year), X402 by Coinbase (May 2025, probably the most popular), KYAPAY by Skyfire/Visa (June last year), AP2 by Google, ACP by OpenAI and Stripe (September last year), TAP by Visa (October), UCP by Google and Shopify (January), ACTP by Alipay, MPP by Stripe and Temple (early this year), AMP by Alipay (April), APOP by UnionPay, an Agent Payment Protocol by OKX, and AgentPay for machines by Mastercard. Jan Curn notes it is really hard to keep track of all the new standards.
some dates flagged as possible numeric errors
AI Engineer · Why AI Agents Still Won't Pay for Anything — Jan Curn, Apify
Creator perspective · MoneyIn May, X402 launched the 'exact' payment scheme, the first for fixed pricing. It then released an 'up to' scheme allowing variable costs, but it didn't work for Apify: nothing prevented the buyer agent from spending the money twice, because the transaction was only settled on blockchain later — so 'up to' wasn't useful for pay-per-use, a disappointment. Apify worked around it using exact plus refund: charge an amount upfront (e.g., $10), and once the work is done and consumption is known (e.g., $2), refund the rest. But that requires the agent to trust the seller — looping back to the trust problem — and involves two blockchain transactions, meaning more fees, so it wasn't ideal. Finally, the batch settlement scheme solved the double-spending issue, and Apify added it about a week before.
AI Engineer · Why AI Agents Still Won't Pay for Anything — Jan Curn, Apify
Creator perspective · MoneyNick Nisi describes a demo progression: the first verification run is memory-only — nothing is stored anywhere, so putting in a wrong email address loses the confirmation. After applying a persistence fix and adding honest copy to the app, the proof can be run again. He frames the result as having 'given your app eyes': the agent can see the app, run checks, change the world of the application, observe the results, and interpret what happened. Running the snack judge command checks whether the confirmation matches what the model saw.
AI Engineer · Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS
Report · MoneyThe speakers describe Snack as a template showing how to set up verification for your own applications, tying into the LLMs working on that application with you, rather than a reusable tool.
AI Engineer · Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS
Report · MoneyWisedocs built a benchmark called MLCR with Artificial Analysis to measure the effectiveness of AI in answering medical claims questions.
AI Engineer · Insurance Claims Are AI's Messiest Data Problem — Denys Linkov, Wisedocs
Report · MoneyThe more complex a claim, the more likely a conflicting diagnosis; a missing referenced x-ray can change the whole trajectory of a case, including whether adequate care was given; records get missed (missed faxes, glitched faxes, changed back-end systems), so processors must ask repeatedly.
AI Engineer · Insurance Claims Are AI's Messiest Data Problem — Denys Linkov, Wisedocs
Creator perspective · MoneyYou only get cross-claim and portfolio capabilities if you can solve the data extraction problem within one claim and then roll it up into your broader insurance portfolio; once good data exists, decisions follow and claims information can be written into a broader system.
AI Engineer · Insurance Claims Are AI's Messiest Data Problem — Denys Linkov, Wisedocs
Creator perspective · MoneyFrom reading traces: an agent running in restricted mode that cannot write files directly is a failure mode that must be fixed; an agent asking permission for things it was told it's not allowed to do is harness leakage; and an agent that munges data from two different web searches into one statistic without checking they're from the same dataset produces confidently wrong output (the Tesla 116% analyst consensus failure).
AI Engineer · How to Know If Your AI Agent Is Any Good: Evals With Phoenix & Claude — Laurie Voss, Arize AI
Report · MoneyLaurie Voss walked through tracing by hand because you should understand what it's doing, but you can also get your coding agent to do all of it. Phoenix has a really good CLI: run its setup command in your project and it will get your coding agent to add tracing to your app. The setup will keep working until your app has tracing installed, so you don't have to figure out how to add the register command yourself, and it works in every language.
AI Engineer · How to Know If Your AI Agent Is Any Good: Evals With Phoenix & Claude — Laurie Voss, Arize AI