head of applied ai ramp.com 💳 investing kahanicapital.com
New York, NY
- Tweets 4,631
- Following 1,340
- Followers 4,693
- Likes 14,455
Jev in Router.com
Here's a quick look into how we're using @typesafeai's Jev here at Router.com.
We're still innovating with this a ton here at @tryramp , so follow along!
2
71
8,545
Ashwin benchmarked Jev as a replacement for LLM-based reranking in Ramp's accounting product, and the results are very promising
Jev matches current accuracy on GPT-5.6 Luna, but decreases tail latency by 10x to 300ms at 3x lower cost
We're excited to get this into production for 70k customers as soon as we stop running into rate limits :)
26
15
180
16,466
Neel moved Ramp’s structured receipt extraction from Gemini 3.5 Flash-Lite to DeepSeek-V4.1-Flash:
- 70% cheaper (~$40k/month saved)
- about the same accuracy
- about the same latency
Running through router.com
23
9
430
58,389
Companies are likely spending millions subsidizing employees’ personal ai use because oai and anthropic can’t figure out how to build a decent account switcher
19
3
303
14,527
This chart is a Rorschach test for how bullish or bearish you are on the AI trade
New from Ramp data: the latest threat to the AI trade. AI companies' revenues are heavily dependent on a small set of customers. 80% of OpenAI and Anthropic's enterprise revenues come from 1% of their customers, and it's not getting better.
This is a level of concentration risk unseen in any other software category we track. The companies in the top 1% skew heavily toward the tech sector and AI products and services.
What happens in a market correction? All these companies are highly correlated, and an increasing share of our economy is invested in them. Especially as we approach blockbuster IPOs for OpenAI and Anthropic.
7
3
61
8,568
Ramp and Router.com are going to integrate seamlessly to help your business understand what work your tokens are producing to measure ROI
DM or email me if you'd like a preview
4
2
93
9,487
Veeral Patel retweeted
The Modern Data Stack is over, long live the Post-AI Data Stack🫡🤖
I wrote about how technological shifts change what a data team can build, the features of post-AI data stacks, what we've built at @tryramp, and what data teams can learn from @nbcsnl.
iandmacomber.com/blog/post-a…
ALT Diagram of the ~2026 post-AI data stack. Every source now sits inside the reachable boundary: unstructured data (Gong calls, emails, docs, code), clickstream, third-party tools, and OLTP databases flow through managed ETL into data stores — Snowflake for structured data, Turbopuffer for unstructured — modeled with dbt, with Reverse ETL out to operations. New in blue: a data agent harness of tools, models, and skills sits between the stores and the interfaces (coworkers, coding agents, Slack bots, AI-native BI tools), fed by company context — semantic layer, lineage, domain docs, activity metadata — through progressive disclosure. An orange feedback loop routes artifacts, analysis, decisions, usage, and evals back into that context.
25
43
457
130,067
The companies of the future will allocate compute like capital: continuously routing it toward high ROI outcomes
Ramp is building the tools to help
Right now, router.com’s traffic is compounding at roughly 34% per weekday. It’s early, but the product is useful and the growth curve is as steep as any I’ve seen.
2
1
82
5,947
Veeral Patel retweeted
Fiduciary duty may not require directors to choose open-weight models, but it should make it increasingly uncomfortable for them to approve material proprietary-model spending without knowing whether open alternatives can perform the same work for less.
Now that credible open-weight alternatives offer material savings, greater data control or reduced vendor concentration, that process should normally include comparative benchmarking and total-cost analysis.
As AI spend becomes material and credible alternatives offer potentially order-of-magnitude savings or materially different security and dependency risks, the duty-of-care expectation of an informed process increasingly supports a documented alternatives analysis.
Today's a record day for open weight share of tokens on Vercel AI Gateway.
Aug 22 (today)
🟦 Open: 62%
🟨 Closed: 38%
Jun 24 (~2 months ago)
🟦 Open: 28.4%
🟨 Closed: 71.6%
This is very likely just the start, because enterprise adoption is still early, and harnesses, CLIs, IDEs, SDKs, etc need to be adapted to be model agnostic.
4
13
66
12,913
Ramp is a "save you time and money" company
router.com is our first step to go beyond fintech
DM me if this is exciting to you- we're actively working on money saving products across the AI stack (router, harness, application layer)
the request-level evidence is the interesting part:
Default baseline: $0.001219
Flex actual: $0.000610
same model, roughly half the cost.
this is the first test that gave me a concrete reason to use Router instead of calling Luna directly.
Nice work, @tryramp , @vral and team!
next thing I’d love: aggregate baseline vs actual spend, flex share, fallback count, and why each tier was selected.
4
4
104
12,886
Veeral Patel retweeted
try Fable in the codex harness with router.com. Honestly it's been game changing for me.
Monitor and control your AI spend on every provider on router.com. Our early users save 40% on average.
Every week, the price-intelligence-latency frontier shifts, and we expect this trend to continue. Tradeoffs between latency, reasoning, cost, service tier, open source and closed source models are shifting constantly.
Router sends every request to the model that's actually best for the task and helps you control what tokens you buy. We benchmark it against real work: ~40% lower cost for the same outputs.
Today we're opening it to everyone.
Two lines of code or just change your base URL. No @tryramp account needed.
Free through 2026, first $26 on us.
Get an API key today at router.com
3
2
21
3,238
We're grateful for the incredible response we’ve gotten since the router.com launch yesterday.
The mission to get more out of AI spend is clearly resonating, and we’re excited to be partnering with @OpenAI to continue doubling down here and make AI even higher ROI for businesses.
the frontier just got more affordable.
we teamed up with @OpenAI: GPT-5.6 Sol is 50% off on router through September 18.
one endpoint, every model, each request routed to the best fit. so every AI dollar goes further.
start saving in a couple minutes at router.com
1
1
61
5,642
Monitor and control your AI spend on every provider on router.com. Our early users save 40% on average.
Every week, the price-intelligence-latency frontier shifts, and we expect this trend to continue. Tradeoffs between latency, reasoning, cost, service tier, open source and closed source models are shifting constantly.
Router sends every request to the model that's actually best for the task and helps you control what tokens you buy. We benchmark it against real work: ~40% lower cost for the same outputs.
Today we're opening it to everyone.
Two lines of code or just change your base URL. No @tryramp account needed.
Free through 2026, first $26 on us.
Get an API key today at router.com
149
109
1,390
1,425,679
Everyone is underestimating NVIDIA's ability to innovate up and down the AI stack
We implemented NVIDIA Switchyard on our internal SWE-Bench and cut costs by 58%
It's live in Ramp Router today - we're actively removing people from the waitlist while we manage load!
We tested NVIDIA NeMo Switchyard’s stage router for coding agents in Ramp SWE-Bench.
Routed agents showed comparable performance to single-model controls while substantially reducing costs and runtime.
5
7
118
19,015
there’s no substitute for being obsessed and making sure every token counts 🫡
There is no substitute for looking at the data. Manually inspecting failure modes is how you discover the shape of the problem.
Great AI products are built by teams who set up their infra + instrumentation to make looking at data easy.
@austospumanto @bcherny @rahulgs
1
1
100
18,487
The company that figures this out is going to save knowledge workers a lot of time and money
What’s the best open source version of Claude Cowork that a) allows you to plug in @OpenRouter or @bittensor b) works with local models and c) is optimized for the average knowledge worker (Hermes and OpenClaw too complex for normies)
Right now Perplexity computer is my best choice (doesn’t do local models yet)
3
1
114
26,196
Veeral Patel retweeted
Set up Ramp Router in ~5 min for my vibe coded workout tracker. Tested about a dozen models on different use cases, and found pretty comparable performance between them for most basic tasks.
The app is now 94% cheaper and running primarily on deepseek-v4-flash
Every week, the price-intelligence-latency frontier shifts, and we expect this trend to continue
Across 100+ use cases in our product, keeping all up to date with the right model is a challenge -
Either we're losing out on intelligence for the dollars we spend, or we're spending too much money for the intelligence the feature needs
Ramp Router lets you benefit immediately without rewriting your application. We cut our LLM costs by 30%, while also making our features smarter and faster.
We built it for Ramp. Now we’re opening it up to everyone.
get access here: ramp.com/router?utm_source=X…
2
2
26
2,290
routers are cool i guess
Happy Wednesday. On today's show:
- @travisk (Atoms)
- @jasonfried (37signals)
- @lqiao (Fireworks)
- @maxhodak_ (Science Corp.)
- @vral (Ramp)
See you on the stream.
4
72
11,557