Recommendation · MoneyWorkOS created their own MCP gateway early in the development of their factory, which they call their context engine — it connects into all of their internal systems and builds system prompts and context around how to navigate the tools and when to use them. In the MCP server's tool description, provide the agent a list of the tables, the content they contain, and guidance like: if you're trying to answer questions about this type of content, write a query for these tables. The MCP server gives the agent orientation to understand how the organization uses its tools, how Linear and Snowflake are organized, and guidance on where to find information. If you're just getting started thinking about a software factory, it is well worth your time to invest in an internal MCP gateway server that is
hedged
AI Engineer · No, That's Not a Software Factory — Ryan Cooke, WorkOS
Recommendation · MoneyRyan Cooke recommends investing in an internal MCP gateway server that connects all internal systems, with descriptions telling the agent how to use those tools and how information is organized in them. It will be useful in many other use cases.
hedged
AI Engineer · No, That's Not a Software Factory — Ryan Cooke, WorkOS
Recommendation · MoneyThe source recommends that in defense, companies should not wait for the government to specify what to build. Instead, they should watch government signals and build in advance using private venture capital, then sell to the government once contracts appear. The traditional defense prime model—'tell me what you want, I'll develop after you give me the money'—is criticized as producing systems that are not technology-forward, developed for 10 years ago, and not valuable to the end soldier. Castelion is cited as an example: it built a reliable hypersonic missile in advance of government contracts using private venture capital, and the government then showed up and bought missiles. If this model repeats, new companies will take substantial market share from legacy primes, leading to better
hedged
The a16z Show · The Startups Rebuilding American Industry | The AD50
Recommendation · MoneyJack Wotherspoon recommends building multi-agent workflows by pointing different agents at the same sandbox so they can write files and build on each other's work. He recommends passing the environment ID from one agent to the next so the new agent reuses the file system but gets a fresh context window and conversation, and having one agent write its output to a file so the next agent can pick up from that input.
AI Engineer · Build Your Own Team of Managed Agents - Jack Wotherspoon, Google DeepMind
Recommendation · MoneySam recommends moving control out of the agent's context or prompt and into the gateway so agents physically cannot perform unauthorized actions — a more sustainable, scalable, and secure approach than agent-internal controls.
AI Engineer · AI Agents Don't Read Your Policy Docs. They Hit Your APIs — Gravitee
Recommendation · MoneyTo scale past hand-auditing, give the agent a scheduled bedtime pass: it reads its own skills, finds stale ones and contradictions, and opens drafts for review — the system sleeps, cleans house, and wakes up sharper.
AI Engineer · Scale the Judgment, Not the Model — Andrew Orobator, Reddit
Recommendation · MoneyOrobator recommends making team judgment explicit in the mechanism that fits: a persona, a lint check, or a commit hook — not just a skill, which is only one implementation. He also suggests finding the judgment the team always asks the same person about as the starting point, and pushing the result org-wide so every team can encode its own taste for its own agents. Spread it by making it legible rather than mandating it.
AI Engineer · Scale the Judgment, Not the Model — Andrew Orobator, Reddit
Recommendation · MoneyTo shift from product to harness engineering, find the judgment your team always asks the same person about and make it explicit — don't just write a skill, which is only one implementation.
AI Engineer · Scale the Judgment, Not the Model — Andrew Orobator, Reddit
Method · MoneyTenev describes how agentic trading accounts are set up and operate: create a separate segregated account and move money into it, keep trade approvals on by default until you want the strategy to run autonomously, then run a daily agent loop that executes your strategy, rebalances the portfolio, and looks for opportunities — optionally connecting alternative data sources like satellite data via the app store. The agent is sandboxed and cannot access other accounts or move money outside the segregated account.
Method, not a verified protocol.
No Priors: Artificial Intelligence | Technology | Startups · Tokenization, Agentic Trading, Harmonic: Robinhood's Mission to Expand Ownership with CEO Vlad Tenev
Recommendation · MoneyThe speaker says it is extremely important to keep testing regression cases, known tasks, and past failures, and that fresh challenge cases at all points go back to setting the most ambitious target. You have to keep pushing your model and make sure it is challenged and has cases where it is not able to perform to expectations.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker identifies adversarial tests as the third part of a holistic evaluation system: find where you can break the workflow and put it into challenging situations. He advises choosing a test for the failures it can reveal — the idea of a test that can fight back.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker identifies two ways to get to an eval that is as holistic as possible: consistently human review your process, and ensure that your customer feedback is part of your eval pipeline. He explains these catch the last 5% edge cases that a model or an existing eval framework might not catch.
hedged
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker advises starting with the job and asking what must happen, what must never happen, and when it should stop, then extending these core questions into other WH questions and first principles to think about the most ambitious outcome. He also endorses aiming at the job you need, not the score you can get.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyFor high-stake applications, the speaker's first recommendation is to define the most ambitious success before designing any test. He suggests assuming a world of AGI: what is your AI agent going to do, or what is the best outcome you can get out of it? Setting this most ambitious goal gives you the scope of what you can go ahead and design further.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker advises continuing human review and making sure the results make sense, and discourages relying on your agents to be the reviewer of your eval at any point, ever. He notes you can fast track your eval process with agents, but make sure you're doing human review.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker advises making automated checks and evals part of your CI pipeline, with eval questions asked to your agent every time they go through a production deployment process. He explains this ensures you're not just taking one eval score at the time you're deciding the model.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker advises continuously testing the workflow at every part of the workflow, discourages assuming things in the workflow or assuming an input and output to be the outcome of anything, and says to make sure your tests are designed for what that element of the workflow does. He notes the objective in that workflow process is to know your tool call ability and tool call accuracy at that point.
AI Engineer · If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com
Recommendation · MoneyThe speaker advises reading real AI outputs, which evals make possible: setting up the right eval tools let Bumble see what real users were receiving and whether the prompt was doing a good job. Bumble's success criteria changed from overall green-versus-red results to examining the long tail of data — whether different segments of users receive feedback well, including whether women respond the same way as men and whether users in different countries respond the same way, since language generated in French may differ from English.
maybe the language generated in French is different from the language generated in English
AI Engineer · 78% Wanted This AI Feature. Almost Nobody Used It - Supriya Subudhi, Bumble
Recommendation · MoneyThe guest recommends enterprises use deterministic controls — sandboxes and human-in-the-loop approval gates — to feel safe deploying computer-use agents, even when the model itself is highly aligned.
The a16z Show · OpenAI: When AI Agents Use Computers Faster Than You
Recommendation · MoneyMasaba Gupta advises making everything on brand and never going off brand, keeping the storytelling and language the same across all brand touch points — she calls it the brand's 'body language'. She stresses that the story created in the head office must permeate all layers of the brand: retail, e-commerce, social media, communication, and all touch points, and that when sitting down with each team, whether creative or retail, she delivers a singular message.
Figma · Config India 2026: What makes the House of Masaba feel like Masaba