Able Varghese

Applied AI for B2B software that makes revenue. AI, SaaS and startup news the day it breaks, with the numbers and what they mean. Founder in Toronto.

[Able Varghese @AbleEmergence]

I track how AI is changing B2B software, using the numbers companies report themselves.

Recent ones: Atlassian's MCP server at over 15M tool calls a day. Stuut running 81.7% of outbound collections with no human. Nous Research saying Hermes was cloned 24M+ times.

Same pattern every time. The revenue goes to the agent that does the work, not the seat that watches it.

[Able Varghese @AbleEmergence]

Three months to five minutes.

That's what CI/CD automation I architected at CIBC did to one deployment process. None of those months went to writing code.

AI coding agents speed up the writing. If your path to production still takes a quarter, they just grow the queue. Fix the pipe first.

[Able Varghese @AbleEmergence]

Rival labs training on Claude's outputs were 5 to 7% of Anthropic's revenue, per the FT. Anthropic cut it back anyway.

Same investor day: Q3 revenue hit $16.8B, up 45% from Q2, and the company said it was profitable on an adjusted basis (FT, Reuters). That distillation spend got cut in Q3 and the quarter still grew 45%.

If you sell an API, some of your biggest accounts are training your replacement. Revenue quality is about to sit next to revenue growth in every AI diligence memo.

[Able Varghese @AbleEmergence]

@mfishbein @a16z @omooretweets Businesses are already a step past paying and into cutting. AngelList says Ramp's token spend tool took 17.5% off its monthly AI bill. Consumers are still deciding whether to pay while CFOs decide how much less to pay.

[Aaron Levie @levie]

Moving from chatting back and forth with agents, to having agents that you spin up in the background to do useful work for you is probably on the order of 100X or 1000X more token consumption than what we’re seeing today.

Most work right now is essentially rate limited by our ability to process and work with information. Moving data between systems. Reviewing a document. Looking through code for security issues. Processing log data to look for even patterns. Looking up information about a prospective client. Almost all of knowledge work is constrained by these steps, and AI has become a huge unlock to that.

So our first step was to use AI to do these individual tasks much faster for us. That was the chatbot phenomenon that we have been experiencing for the past 3+ years. But we can still then only augment the work that we remain in the loop for, and we’re highly limited by our ability to parallelize this work by our own bandwidth and context.

But where we’re going is quite different. As AI models continue to improve at incredible rates, we can hand off more and more work to the agent. First it’s letting the agent run with a harder problem. Then it’s giving a few agents multiple parts of a task at once. But we’re not far off from being able to throw an arbitrary number of agents at a problem and just say “come back when you’ve accomplished this goal”.

Now imagine that agent swarm for any problem we can think of. Protecting our codebases, data, and systems. QAing every change we make and make sure all our infrastructure is robust and does what it should. Looking through every document for potential risks. Auditing this company with vastly more data reviewed. And on and on. Most of this isn’t even replacing existing work we do, it’s all net new.

These agents will get deployed as swarms by people, by workflows, by triggers, and by other agents. And none of this would be possible if the models weren’t getting better, and if the cost of AI didn’t make it affordable. Both of those trends though are playing out. Now the next frontier is just having enough compute. But this is where it’s heading.

[Able Varghese @AbleEmergence]

@jbahrdestefano @tryramp @business Card volume is running even hotter than revenue: $275B in annualized purchase volume, up 175% year over year per the release. That gap is pricing room Ramp hasn't touched yet.

[Able Varghese @AbleEmergence]

Ramp just raised $1.85B at a $60B valuation, and half the launches it listed today are about AI tokens, not cards.

Dragoneer led. Four months ago it was $44B. Ramp says it's past $2B in annualized revenue with positive cash flow, and $100K+ customers are up 113% year over year. New on the list: token spend controls across model providers, plus a router it says cuts inference costs up to 40%.

Finance is about to own the model budget. Whoever sits in that approval flow gets a say in which AI vendors keep the account.

[Able Varghese @AbleEmergence]

Ramp just raised $1.85B at a $60B valuation, and half the launches it listed today are about AI tokens, not cards.

Dragoneer led. Four months ago it was $44B. Ramp says it's past $2B in annualized revenue with positive cash flow, and $100K+ customers are up 113% year over year. New on the list: token spend controls across model providers, plus a router it says cuts inference costs up to 40%.

Finance is about to own the model budget. Whoever sits in that approval flow gets a say in which AI vendors keep the account.

[Able Varghese @AbleEmergence]

Slack only has real loyalty among developers, says Bessemer's Adam Fisher. Zenzap just raised $44M to go after everyone else.

Bessemer, Lightspeed and The Garage are in. Founder Guy Weiss bootstrapped it out of Bessemer's office before taking a dollar. FedEx, Dollar General, Taco Bell and Starbucks are listed clients. Paid plans start at $51 a month for up to 20 users, and AI agents bill on credits.

Frontline teams already run shifts on WhatsApp. Their first work agent shows up wherever IT lets that chat move.

[Kyle Harrison @kwharrison13]

The age of cloud computing as a triopoly is over.

There was a bygone era where MSFT got sued for having a monopoly in software; that would be unimaginable in an age of SaaS.

The same will happen in compute. Most AI companies will have an incentive to be vertically integrated.

[Flo Crivello @Altimor]

We just gave Lindy its own phone number! You can call it on your morning commute, it can call you when something urgent comes up, and you can let it act as your receptionist and have other people call it to e.g. find time with you.

Available today!

[Able Varghese @AbleEmergence]

@jbahrdestefano @LedgerAPI @FirstMarkCap @BoxGroup Shipping an MCP server on top of the ledger is the tell. Once agents call the books directly, reconciliation stops being a month-end job and becomes a gate on every payment.

[Able Varghese @AbleEmergence]

Fragment's whole bet: an AI agent shouldn't be able to move company money until the books balance.

FirstMark and BoxGroup co-led a $12M Series A that closed in mid-2025, announced today. Stripe led the $11M seed before it. Every move gets a double-entry check before the payment clears. Roughly 50 to 60 paying customers, Ramp, BILL and Whop among them.

Agents that can pay got this year's headlines. The ledger that can refuse is what finance teams sign for.

[Able Varghese @AbleEmergence]

Pluto came out of stealth with $42M to watch what AI agents actually do inside the company, not just what users click.

TLV led an $8.5M seed and Jump led a $34M Series A. Selling since April, already at $2M ARR across more than 30 enterprises. About 60% of the risks they catch sit in tools vibe-coded by non-developers.

User behavior security was built for people. The next budget line pays for agent behavior.

[elvis @omarsar0]

Another great application of System One models.

I think much of what we call hallucination in web agents comes from misreading the page.

When a page is flattened to text, a small ✗ in a pricing table disappears, and the agent answers incorrectly with full confidence.

Expand Fetch reads the rendered page, returns the underlying data as clean JSON, and keeps a replay of exactly what your agent saw.

[Nathan Lambert @natolambert]

Very timely appearance on the a16z podcast with @lishali88, to try to understand how open models play into RSI dynamics, who captures value of open models, wrangling agent swarms, and why I'm building Trillium Labs.

[Modal @modal]

Today we're announcing that Modal has crossed $1B in annualized run rate, as well as raised a $750M Series D at a $15.75B valuation, led by @Accel.

Since our Series C, usage has tripled and we’ve grown more than 15x year-over-year.

We are deeply grateful to the developers and companies building on Modal. There's still so much more to build, and we can't think of a better time to be building software.

[Able Varghese @AbleEmergence]

@charles_irl Skating to where inference demand is going is the only infra thesis that survives a 15x usage year. Pricing last quarter's GPU map is how you miss Modal-class growth.

[Able Varghese @AbleEmergence]

Scimplify just turned a 12 to 18 month specialty chemicals cycle into 2 months, and hit about $200M ARR doing it.

Hitachi Ventures led a $25M round at a roughly $250M post. US run-rate went from about $10M to about $60M in a year. AI agents plan the chemistry; more than 500 partner plants ship it.

China concentration is the constraint. Speed plus local capacity is the product.

[Able Varghese @AbleEmergence]

Anthropic's merges went from 2,500 a day in May to over 30,000 in September. Git couldn't keep up.

Seven engineers and Claude built Scion, an in-house source control system, and migrated the monorepo in five months. Checkouts that took minutes now take seconds. Infrastructure CPU and memory dropped about 200x versus the old git mirror fleet.

When agents write the code, the bottleneck moves to tools built for humans. Teams that keep human-scale infra will feel the queue before they feel the model.

[Jerry Liu @jerryjliu0]

Benchmarks are becoming the core competency of any company.

Our company mission is to build the world’s best models for document processing. This also means that we have to build the world’s best benchmarks for doc parsing and extraction.

It’s been incredible to see the adoption of ParseBench the past few months. We have 26k+ monthly downloads on Hugging Face, 147+ models benchmarked, and has been cited by other labs in the space (Cohere, Mistral, etc.).

We use this as a benchmark to push our own internal capabilities within LlamaParse, and also to track the advancement of recent frontier/open-weight models (most recently, Gemini 4 Argon and Muse Spark/Wonder).

We’re happy to help apply this to any company looking at evaluating document processing solutions. Even if you don’t use our docs, you may find our metrics for evaluating various dimensions (e.g. tables, reading order, layout) useful.

Check out ParseBench: https://t.co/PWczfhp0OX

Hugging Face: https://t.co/meFLEzzgD8