vas natarajan

partner @ Accel. head cheerleader @ gamma, lovable, radar, ironclad, transcend, permitflow. formerly: frame (ADBE), spoke (OKTA), segment (TWLO), deepmap (NVDA)

[vas natarajan @vas]

What a moment, @cursor_ai! Congrats to you! How often do you see a company's pace of innovation inflect again after an acquisition announcement... this team is absolutely cooking.

Side note, this will go down as one of the all-time greats @Accel and @Mkclements @agbraccia. To catch a breakout company and stay convicted at every fundraise along a never-before-seen trajectory ... one of the hardest parts of our job is to let your mind explore outcomes that break every internal model. Bravo!

[Miles Clements @Mkclements]

Congratulations and thank you @mntruell and the entire @cursor_ai team.

Grok Bot and Grok 4.6 are just the faintest preview of what's to come from this extraordinary group of builders.

It is so, so early.
https://t.co/cMHbwBKyMo

[vas natarajan @vas]

Lovable blew past vibe-coding a long time ago. The typical customer today is very deliberate about telling their story, serving their customers, and scaling their business etc. They want a platform to meet their ambition -- and @Lovable is their AI-native infrastructure.

[WSJ Tech @WSJTech]

Artificial-intelligence startup Lovable has raised fresh funds at a $13.3 billion valuation, highlighting investor enthusiasm for vibe-coding tools https://t.co/XVwuDcyqY1

[vas natarajan @vas]

.@Permitflow (YC22) one of the faster-growing under-the-radar applied AI companies doing just that-- reasons about what *can* be built (local building regs/zoning laws) -- and connects that with complex knowledge work happening across architects, engineers, lawyers, designers, etc to iterate on approvals and licensure.

[Paul Graham @paulg]

Now that AI has compressed the design phase for so many products, there is going to be more demand for technologies that compress the construction phase. After we get those, the final frontier will be regulation.

[Paul Graham @paulg]

Now that AI has compressed the design phase for so many products, there is going to be more demand for technologies that compress the construction phase. After we get those, the final frontier will be regulation.

[vas natarajan @vas]

@theo Dude. You are cooking!

[SemiAnalysis @SemiAnalysis_]

CRACKED ALERT: The startup powering @sgl_project, @radixark, has some of the most hardcore engineers in inference. They power production at @xai and at many Chinese labs.

[vas natarajan @vas]

I largely loiter near the SPC door to find great founders coming and going. @adityaag and team have an uncanny ability to attract deeply curious technical talent willing to go through the ideation grind. Gamma, Transcend, Nuance Labs, and many more @Accel-backed alums who sing the praises of their time at SPC.

[Aditya Agarwal @adityaag]

Today we’re announcing SPC Fund IV: $575M to help founders find their life’s work.

A decade of guiding founders through -1 to 0 brings @SPC to $2B in AUM and, more importantly, a 1200 member community of the world’s most talented technologists. We strongly believe that:

→ The person comes before the idea. At -1, generational founders are often busy reading, prototyping, discarding, and earning the right to their own conviction.
→ Ambition is social. The people you surround yourself with in the early days play a big part in inflecting your ambition.
→ Patience compounds. It is better to work on something you have deep conviction in rather than simply sprint at the first idea.

The quiet months we protect at the beginning have produced three funds in the top 10% of their vintages, and companies like @baseten, @GammaApp, @render, @GoodfireAI, @LumaLabsAI, and @profound. We are proud to have worked with all these founders in their -1 phase and beyond.

While we defined -1 to 0, we're no longer bound by it. Fund IV allows us to partner with companies well beyond launch.

We're doing this now because we've never seen a wider gap between what's possible and what's being attempted. Intelligence is abundant. The cost of trying ideas keeps falling, and problems that seemed insurmountable a few years ago are quickly becoming solvable.

To our members, to the founders who trusted us before anyone else did, and to the LPs who had conviction in our own -1 journey: thank you. This milestone belongs to you as much as to us.

To the builders looking to find your life’s work: let’s get started.

[vas natarajan @vas]

So much interesting work happening to pull different domains of knowledge work up the verifiability ladder -- so AI can more clearly operate against them. Law, tax, medicine, finance, compliance, construction. Manufacture a verifier where none exist so machines can tell right from wrong.

[vas natarajan @vas]

@jayneil @stephenhaney @Accel @ICONIQCapital I am available!

[vas natarajan @vas]

@llmbenchatkapua Agree price parity shouldn't outright imply all tasks are solved similarly, though that's why the cascade will exist. Maybe weighing operational simplicity AND task-fit changes the ranking on occasion. I'd just argue the former is a stronger consideration than most think.

[vas natarajan @vas]

The open v. closed debate framed around cost misses the broader enterprise TCO picture. Yes, license to an open-weight model is $0, but operating a model means owning everything around it. Closed often wins on operational simplicity (see: plenty of prior-gen infrastructure and API businesses). Enterprises pay to concentrate support, development, and implementation in a single vendor's hands.

That gap is closing though — you can get a near "open-model-closed-ops" experience on the inference clouds.

Still, worth thinking through the cascade of workloads across three paths from buyer's POV:

1. Frontier proprietary — hardest reasoning tasks
2. Prior-gen proprietary — repeatable, nuanced work that doesn't need frontier reasoning.
3. Open-weights on an inference cloud — bulk, batch-y workloads where the boundaries are well defined

The gush of tokens across each path isn't slowing down. Labs will want to tighten the experience between 1 and 2. And if there's price parity between a frontier open-weight model (Kimi K3) and a prior-gen proprietary model (Sonnet 4.6) — the cost argument falls away, and TCO/buying simplicity starts to dictate instead.

Caveat caveat caveat. This is about enterprise buyers. I think a lot of software companies have already made their choice (open).

[vas natarajan @vas]

Rich Paul, the Kalshi whale.

[vas natarajan @vas]

Two kinds of applied AI we see.

Automation-intelligence climbs the same hill faster — bulk repeat tasks, coded away. Real, but the ROI has a ceiling and unit value ($-per-task) can degrade quickly.

Augmentation-intelligence finds a taller hill you couldn't see — it broadens the solution space entirely. High perceived ROI.

But the latter lacks cheap/fast verifiable rewards which also makes it more difficult. There may be proxy rewards over time (i.e. see AI drug discovery startup finally getting FDA approved).

The former can more easily point to "task complete" (i.e. AI help desk sees customer ticket resolved) so the loops are tighter.

[vas natarajan @vas]

The opposite of product insecure -- you gladly build paths out of your product, but use that as motivation to be so good that no one takes the option. @Lovable

[Tyler Bruno @tylerbruno05]

people keep saying you can't leave Lovable:

"you can't get your code out" → two-way git sync, full history, runs anywhere

"your data is trapped" → real postgres export, migrations in your repo. or bring your own supabase from day one

"you have to host with them" → frontend on netlify, cloudflare, wherever. or move the backend and keep the frontend. mix and match

why would you though

[vas natarajan @vas]

Models frozen on historic global data can misjudge local truth. Agents climb the steepest hill in front of them, but it may not be the one the user wants them on. Now the interesting product frontier isn't feeding models fresher facts — it's drawing the right objective out of the user and packaging it up for AI. That's a super complex pathway to solve, but awesome work being done there @noah_weiss @nuance_ai @thinkymachines

[Thinking Machines @thinkymachines]

We're building AI that people and organizations can shape and make their own. AI should extend our will and judgment instead of neglecting it; enabling that is the technical challenge we are working to solve.
https://t.co/Bi558y4vqD

[vas natarajan @vas]

Agents can replace entire application stacks, but in many industries they're forced to be guests in existing systems. The @useagave team spent years reasoning about legacy application ERP and financial software -- the data models, the integration paths. Now their agents power the operations of some of the most critical commercial and infrastructure projects in the country.

Growing like a (ahem) weed, profitable. And an incredible, hard-to-replicate moat.

[Accel @Accel]

Many of today's most interesting AI companies are working in the toughest possible conditions, with complex data, systems built decades ago, and almost no margin for error. @useagave is one of them.

Tom Reno, John Zucchi, and Pooria Azimi are building for a construction industry that is still running on incumbent software, where every job is its own business and few processes have caught up with the modern world.

After leading Agave’s seed round, we're deepening the partnership by leading their Series A. Read more from Accel's @Vas Natarajan ⬇️ https://t.co/vl9Usv5PJu

[vas natarajan @vas]

Building a warm, trusted relationship with founders while simultaneously isolating/ tackling key diligence questions on a company -- that's the spell our late-stage team has mastered. Our IP is less in "a playbook", but in developing partners who can wield that delicate balance in today's compressed timelines.

That's how you earn an outsized share of the best breakouts in each vintage. Proud partner of this team @Accel 🦾

[Molly O’Shea @MollySOShea]

NEW: Inside Accel's Growth Investing Strategy

An exclusive look at 40 years of legendary deals

From a 10% stake in Facebook to:
› Nebius (PIPE, up ~13X)
› Cursor (now a $60B SpaceX acq)
› Cyera (highest-valued private security co)

Portfolio includes Anthropic, Spotify, Lovable, Supabase, Vercel, Scale, Decagon, UI Path, Crowdstrike, Linear.

With Growth Partners Arun Mathew, Miles Clements, & Matt Weigand, we cover:

› Agent economy & real enterprise adoption
› Inference buildout and infra as the rate limiter
› Token maxing & where spend goes next
› 3 trillion dollar IPOs & the race to $10T companies
› Why security exploded after Mythos

[vas natarajan @vas]

One of the more interesting parts of applied AI is defining the goal to iterate against. Code compiles — its a machine-verified check an agent can goal-seek towards. But most verticals lack that check; the goal is an affirmative email, a signature, a nod, sometimes undefined altogether. Great applied AI products seem to find that goal and pull it into the loop.