Creator perspective · MoneyPhilipp Schmid summarizes that the full AI co-worker in one service consists of: the brain (Gemini), the computer (managed agents), skills (same as local), tools, MCP servers for external services and data providers, a secure way to provide and rotate credentials, hooks for deterministic logic protection, and triggers for proactive integration.
AI Engineer · 7 Things Your AI Coworker Needs - Philipp Schmid, Google DeepMind
Creator perspective · MoneyPhilipp Schmid explains that the benefit of the credential proxy approach is that there's no way for a prompt injection to access the credentials. With the network proxy approach there is no way for a prompt injection to steal credentials, because the proxy only swaps them for the domains the user defines. A malicious prompt injection can ask the agent to print all values in a .env file, and the agent might go there and print all of the keys. With per-domain credential swapping in the proxy, a malicious MCP server cannot steal your Gmail token. The network proxy approach makes it much easier to trust the agent and to integrate MCP servers the agent can use.
hedged
AI Engineer · 7 Things Your AI Coworker Needs - Philipp Schmid, Google DeepMind
Creator perspective · MoneyPhilipp Schmid explains that credentials are super hard and super easy to get wrong with agents, which is why people are very careful about using agents because things could go wrong very easily. With managed agents, credentials are separated via an external network proxy: instead of providing the agent the credentials inside the sandbox, you create credentials externally. External credentials are write-only: once written, there's no way to retrieve the credentials, you can only override them, so there's no possibility of leaking. Managed agents support three different ways of credentials — bearer tokens for normal API calls, OAuth for consumer-oriented MCP servers and tools, and environment variables for CLIs and other features — making it secure and isolated for your agent to use them.
AI Engineer · 7 Things Your AI Coworker Needs - Philipp Schmid, Google DeepMind
Principle · MoneyThe enablement program outline breaks work step by step into little milestones so nothing is missed, and the team builds out every task with milestones in a follow-up meeting.
30 Minutes to President's Club | No-Nonsense Sales · #623: My Sales Enablement System that 3x'd My Team's Quota Attainment
Report · MoneyRaymond Weitekamp plays Factorio with a recursive decision model that identifies which line/group has a problem, pinpoints the faulty element (an inserter), and runs over to fix and replace it. In the demo there is no large language model and nothing generative at all. His goal is to have the RDM teach him how to play Factorio, since he never really played before. He notes the demo lacked pathfinding composition of recursion, so it ran the long way around to the next target. Watching it without knowing the internals, one would assume some kind of reasoning model is needed, yet it works without any generative language model. He endorses Factorio as a very open-ended way of evaluating and benchmarking systems.
hedged
AI Engineer · Your Agent Doesn't Need an LLM for Every Decision - Raymond Weitekamp, OpenProse
Report · MoneyVercel deployed a lead qualification agent with a human in the loop in August of last year, reducing the SDR team from 10 people to 1 over six weeks. The agent now runs 24/7 with faster speed to lead and human-equivalent quality metrics, and one person runs lead qualification in the US with about 20% of a person covering Europe and APAC.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 885: Going from 10 to 1 SDRs. How We Deployed AI Across our Sales Org with Vercel's COO
Report · MoneyVercel's head of data science built a semantic layer on top of a model of revenue, enriched with third-party and first-party signals, and every agent runs on top of it. This knowledge phase is what makes Deal 1's analysis actionable rather than just interesting, and what makes the Playbook Platform contact the right companies with the right message. Without it, agents hallucinate or give generic answers; with it, agents are grounded in Vercel's data.
The Official SaaStr Podcast: SaaS | Founders | Investors · SaaStr 885: Going from 10 to 1 SDRs. How We Deployed AI Across our Sales Org with Vercel's COO
Report · MoneyAt WP Engine, fewer than 10% of customers were causing a million dollars a year in profit to be gone, dragging profit margin from 80% to 55%. The top 10% of customers represented about 40% of revenue. Cohen notes a 55% vs 80% margin is the difference between investors not treating you as a software company and being best in class. He showed all the data for the 7,000 WP Engine customers in the book.
Startups For the Rest of Us · Episode 857 | Finding Hidden Multipliers in Your Business (with Jason Cohen)
Report · MoneyJason Cohen recounts that at WP Engine, unprofitable customers were causing a million dollars a year in profit to disappear, dragging the pro-growth profit margin from 80% down to 55%, and that the top 10% of customers accounted for 40% of revenue.
Startups For the Rest of Us · Episode 857 | Finding Hidden Multipliers in Your Business (with Jason Cohen)
Creator perspective · MindHuberman states, citing the data, that two alcoholic drinks per week is probably the upper limit for an adult non-alcoholic. He also observes that for some people, drinking no alcohol at all is easier to stick to than drinking some.
'probably the upper limit' per the data, as he frames it
The a16z Show · Andrew Huberman’s Three Rules for Better Health and Focus
Report · MoneyZachary Blumenfeld describes the execution of his compiled skill: an execution checks hard policy rules first, then routing, then a decision model, then report writing with a linting loop. The decision model is given explicit inputs — whether there is exposure material and the outcome of approving or denying — and a router escalates to human review when the model does not auto-approve or deny.
AI Engineer · Make Your AI Agents Predictable With Graph Engineering - Zachary Blumenfeld, Neo4j
Principle · MoneyAWS micro VMs have an eight-hour session cap, which conflicts with long-running multi-agent workloads like Factory's missions product, which can run for multiple days of multi-agent orchestration. Nikhil Harithas endorses the pattern of saving the sandbox with a snapshot, rehydrating a new one, and swapping it in place as if nothing happened when a workload exceeds a session cap.
AI Engineer · Your AI Agents Shouldn't Care Where They Run - Nikhil Harithas, Factory
Report · MoneyBlackstone, a trillion-dollar financial institution with heavy regulation and strong opinions about non-deterministic software, could not run agents on laptops. It required bring-your-own compute, bring-your-own inference, its own identity, and single-tenant deployments everywhere.
AI Engineer · Your AI Agents Shouldn't Care Where They Run - Nikhil Harithas, Factory
Creator perspective · MoneyKatie Givens names her three biggest rocks for the first 30 days: the listening tour, focus groups, and the revenue velocity model.
30 Minutes to President's Club | No-Nonsense Sales · #622: Sales Enablement Playbook: How To Win Your First 90 Days
Creator perspective · MoneyKatie Givens describes a 90/10 time split in the first 30 days: most time listening and learning, with a small slice spent on quick wins. She had planned to just listen and learn for 30 days but adjusted to include quick wins.
30 Minutes to President's Club | No-Nonsense Sales · #622: Sales Enablement Playbook: How To Win Your First 90 Days
Creator perspective · MoneyKatie Givens' listening tour means sitting down with the most important people — those sales works with most and she'll work with most. She would prioritize stakeholders that interact with sales the most (legal, CS, implementations, SDRs, and their leaders), and also meet with HR to understand performance management and cases on the sales team, and with product to understand the conversation stream between market feedback and product decisions.
30 Minutes to President's Club | No-Nonsense Sales · #622: Sales Enablement Playbook: How To Win Your First 90 Days
Creator perspective · MoneyKatie Givens describes her listening tour approach: people are typically very upfront, she looks at data to identify the most consistent performers and schedules separate one-on-ones with them, and she reaches out to managers to run focus groups, asking which reps to meet with — new reps who just went through onboarding, tenured reps, and anyone who is highly opinionated.
30 Minutes to President's Club | No-Nonsense Sales · #622: Sales Enablement Playbook: How To Win Your First 90 Days
Principle · MoneyYou need to make sure you cover every route in a skill when detonating it, because any one of them can be malicious; Zenity created an operator that reads through the skill and tries to analyze the different trajectories that should be invoked.
AI Engineer · You Can't Tell What an Agent Skill Will Do by Reading It - Roey Ben Chaim, Zenity
Creator perspective · MoneyAn observed skill trajectory led through cloud and Kubernetes configuration, contacted the identity service, and probed the instance metadata endpoint — the same pattern infostealers use when scanning for secret and identity folders.
AI Engineer · You Can't Tell What an Agent Skill Will Do by Reading It - Roey Ben Chaim, Zenity
ReportJacob Lauritzen reports the billing model is changing: a bunch of stuff is already being priced on a project fee rather than an hours fee, particularly in the US, and lower-end legal work like small reviews is being priced on a flat rate or done for free.
Latent Space · Building Legal Agents When You Can't See the Data - Jacob Lauritzen, CTO of Legora