James Tamplin

MCP is not an integration strategy. We raised $15M from Bessemer to prove it. If you are building an enterprise agent that needs to connect with your customers’ critical systems, simply integrating with the Salesforce MCP isn’t going to cut it. 1. MCP isn’t going to help you backfill millions of transactions from your customer’s NetSuite, or millions of accounts from your customer’s HubSpot. For that, you need robust sync infrastructure. 2. Enterprise customers do not use tools out-of-the-box. Their CRMs, ERPs, and all the legacy systems of record have custom objects, custom fields, and bespoke workflows. MCP will not help you navigate that complexity. 3. MCP is built for real-time querying. Some work is better done async, especially when you are dealing with strict rate limits that many SaaS products have. You need async workloads and smart retry logic. Ampersand is the integration infrastructure for deeply connecting with your customers’ systems of record. Because the SaaSpocalypse didn’t happen. The world still depends on CRMs, ERPs, HRISs, and ITSMs.

1/ Stripe has signed an agreement to acquire OpenRouter. OpenRouter will continue to operate as it is: same name, same product, same roadmap, same mission. But now, we will do it faster, and with Stripe's unparalleled excellence and reach.

One more fun slide from our pitch deck.

A lot to like here, some mechanics to work out, and a few parts to actively build guardrails around. The biggest takeaway is that it is genuinely encouraging to see almost all frontier companies now coming to the table and calling for governance. As I was recently reminded, the best time to plant a tree was 10 years ago. The second-best time is today. I will point out a few things. On the mechanics: if this proposal picks up steam, you will start hearing a lot about the "non-delegation doctrine." The TLDR is that there are real limits on what a self-regulatory body inside the federal government can actually do. Courts are already circling FINRA, the exact model Demis cites, for far less. Whatever department houses this body will have to closely oversee its rulemaking, and in practice, that can make the whole thing collapse into an advisory committee. The way out is to separate the jobs. Government sets the outcomes and keeps the enforcement power. Accredited private evaluators do the measurement. Measurement can be delegated. Enforcement cannot. Now for the part Fathom likes most: the proposal calls for government and industry to help establish an ecosystem of independent evaluators. That is needed. But it comes with an obvious risk of capture. If the evaluators are funded by and answerable to the labs they test, we have rebuilt the credit rating agencies. And to be clear, that failure was never about competition. It was government-protected status, issuer-pays, and no consequences. Very quickly within this ecosystem, evaluators have to compete for government accreditation, not lab favor, and lose that accreditation when their measurements fail. Think UL, not Moody's. Again, I thank @demishassabis for putting this forward. There is a lot of policy ground to work with here.

Grok is maximally truthful

GPT-5.5 lies constantly, Grok 4.20 doesn’t. We made a simulation to see if AI lies when the stakes are life & death.🧵

After 4 years, we’re announcing Instant 1.0. Instant is the best backend for AI-coded apps. Let us tell you why.

In 2013, Facebook bought Parse -- Firebase's #1 competitor at the time. We know how that ended... Now, Facebook just bought @ManusAI, @TaskletAI's #1 competitor. Lets see what happens... 😉