Breaking
Kognitos raises $25M Series B led by Prosperity7 Ventures Named HFS Research Hot Tech 2026 for deterministic AI automation Gold Globee Winner - Neuro-Symbolic AI Platform, 2026 ~$85M raised across seed, Series A and Series B Founder Binny Gill was CTO at Nutanix Kognitos raises $25M Series B led by Prosperity7 Ventures Named HFS Research Hot Tech 2026 for deterministic AI automation Gold Globee Winner - Neuro-Symbolic AI Platform, 2026 ~$85M raised across seed, Series A and Series B Founder Binny Gill was CTO at Nutanix
Company Dossier · Enterprise AI

Kognitos wants you to write software in plain English.

A Mountain View startup built a neurosymbolic engine so the machine understands your sentence - and then refuses to make anything up.

Founded 2020 Mountain View, CA Neurosymbolic AI ~71 Employees
Kognitos brand card: the yellow-green Kognitos monogram on a dark field with the line Hallucination-Free Agentic AI for Business Transformation.

The wordless part of the pitch: a single monogram, a dark field, and one claim - "hallucination-free." No people, no product screenshot. Just a company insisting, quietly, that it will not lie to you.

Share → LinkedIn Twitter / X Facebook Instagram
Filed under: AI · Enterprise · Automation The Business Desk · Approx. 1,700 words

The company that made "don't hallucinate" the product

Here is a slightly funny thing about enterprise software. For roughly a decade, the automation industry sold companies robots - "RPA," robotic process automation - that were not robots at all. They were scripts that clicked buttons on a screen, and they broke the moment somebody moved the button. You paid a lot of money for a bot that watched your screen the way a very literal intern might, and then you paid a lot more money to a developer every time the intern got confused.

Kognitos, founded in 2020 in Mountain View, looked at this and proposed something that sounds obvious and is technically hard: what if you just told the computer what you wanted, in English, and it did that? Not "drag these eleven nodes into a flowchart." Not "learn our proprietary studio in 80 hours." Just - write the instruction the way you would write it to a coworker, and let the machine run it.

The catch, of course, is that machines that understand English tend to be large language models, and large language models have a well-documented habit of confidently inventing things. If your automation processes invoices, "confidently inventing things" is not a charming quirk. It is a chargeback. So Kognitos's actual pitch is narrower and more interesting than "AI that talks." It is: an AI that understands your sentence, and then executes it with the boring, literal reliability of a rules engine that cannot improvise.

"A future where all business applications will be written in English, where machines understand human language natively." - Kognitos, on its own mission

Neurosymbolic, or: two brains in one box

The technical name for this is neurosymbolic AI, and it is having a small renaissance. The "neuro" part is the neural network - the LLM that reads your plain-English instruction and figures out what you mean. The "symbolic" part is the older, unfashionable technology: a deterministic engine that takes rules and follows them exactly, every time, with no creative liberties. Kognitos's patented trick is to keep these two jobs in separate boxes. The language model interprets. The symbolic engine executes. The interpreter is allowed to be clever; the executor is not allowed to be anything but literal.

This is why the company can say the words "hallucination-free" with a straight face. It is not claiming the LLM never gets confused. It is claiming that the part of the system doing the actual work - moving the invoice, updating the record, flagging the exception - runs on rules that were written down and can be read back, in English, after the fact. The automation is also its own audit trail. You can hand the log to a compliance officer, and the compliance officer can read it, because it is just sentences.

The product name for all this is Koncierge, and the branding leans hard into the letter K, which is either endearing or slightly much depending on your tolerance for startup naming. Under the hood it connects to more than 130 enterprise systems, so the plain-English instructions actually reach the software where the work lives.

What you can actually do with it

The honest answer is: unglamorous, expensive back-office work. Invoice processing. Order handling. Reconciliations. Reading a messy PDF that a supplier emailed and turning it into a clean record. Handling the exception when the messy PDF is missing a field - which, in the real world, is most of the time. Kognitos says its platform auto-resolves better than 90% of exceptions, which is the sort of number that sounds like marketing until you remember that exceptions are precisely the part legacy automation could never handle, because a script that clicks buttons has no idea what to do when reality deviates from the script.

The people meant to build these automations are not developers. That is the whole point. Kognitos claims a business user can get proficient in 8 to 10 hours, versus the 80 to 120+ hours it cites for learning a traditional automation studio. Whether those exact numbers hold up in your finance department is a fair question. The direction is the interesting part: the person who understands the process is now the person who can automate it, without filing a ticket and waiting three sprints.

"Using Koncierge, anyone can describe what they want automated, and their automation is generated in auditable English." - From the company's product materials

The founder did this before, sort of

Kognitos was founded by Binny Gill, along with co-founder Nicholas Brigham Adams. Gill's previous job is the kind of line that makes investors return your call: he was CTO at Nutanix, the enterprise infrastructure company he helped scale to roughly $1.5 billion in revenue. Then he left to bet on a sentence.

You can read the thesis as a natural sequel. Nutanix was about hiding infrastructure complexity so companies could just run their software. Kognitos is about hiding software complexity so people can just describe their work. Both are bets that the abstraction most people should have to think about is much higher than the one the industry currently forces on them. Gill's version of the future is one where interacting with a computer is as easy as talking to a human - which is a big claim, and also more or less what everyone in AI is now racing toward from different directions.

The money, briefly

The funding history is tidy and fast. A $6.75 million seed round in early 2023, led by Clear Ventures. A $20 million Series A later that year, led by Khosla Ventures. Then a $25 million Series B in June 2025, led by Prosperity7 Ventures, with Khosla, Wipro Ventures, Engineering Capital, Dentsu Ventures and Alumni Ventures along for it. That is roughly $85 million total in under five years, earmarked to push into banking and healthcare - two industries that care a great deal about audit trails and very little about vibes.

The recognition has followed the capital. Kognitos was named an HFS Research Hot Tech 2026 company for deterministic AI automation, took a Gold Globee as best Neuro-Symbolic AI Platform, and picked up a 2026 AI Breakthrough award for natural language understanding. Awards are awards. But the through-line is consistent: the industry keeps categorizing Kognitos by the thing it does differently, which is the deterministic, symbolic half of the sandwich.

The real competitor is the status quo

On paper, Kognitos competes with UiPath, Automation Anywhere and Blue Prism - the incumbents of the RPA era. The company happily publishes comparisons claiming faster deployment, roughly 12x lower maintenance, and far fewer help-desk tickets than legacy bots. Take vendor benchmarks with the usual grain of salt. But the more honest framing is that Kognitos is not really competing with a product. It is competing with the enormous installed base of brittle scripts and the developers paid to keep them alive. Replacing that is less a sale than a habit change, which is slow, which is also why the round is aimed at giving the company runway to wait it out.

There is a version of the near future where "write it in English" becomes the default way ordinary people build automation, and a version where regulated enterprises decide that trusting any AI unsupervised is a bridge too far. Kognitos has planted its flag on the argument that the second fear is exactly what its architecture solves - keep the creative part and the reliable part in separate boxes, and you can have the language without the lying. It is a clean idea. The next few years are about whether clean ideas survive contact with accounts payable.

2020
Founded
~$85M
Total Raised
130+
Integrations
90%+
Exceptions Auto-Resolved*

How the two-brain trick works

Interpret · then execute · literally
STEP 01

You write English

A business user describes the process in a plain sentence - no flowcharts, no code.

STEP 02

The neural half reads it

An LLM interprets intent and turns the sentence into structured, auditable steps.

STEP 03

The symbolic half runs it

A deterministic engine executes the rules exactly as written - the same way, every time.

# What "English as code" looks like
Get the invoices from the shared inbox.
For each invoice, read the vendor and the amount.
If the amount is over 10,000 dollars, ask a manager to approve it.
Otherwise, record it in the ledger and file the document.
# ...and the log reads back in the same plain English.
Seed to Series B

A five-year run

2020

Founded

Binny Gill, fresh off scaling Nutanix, co-founds Kognitos with Nicholas Brigham Adams to make machines understand human language natively.

FEB 2023

$6.75M Seed

Clear Ventures leads, with Engineering Capital and Wipro Ventures, to bring generative AI to enterprise automation.

NOV 2023

$20M Series A

Khosla Ventures leads, pushing total venture capital raised to about $30M.

JUN 2025

$25M Series B

Prosperity7 Ventures leads the neurosymbolic-platform round, targeting banking and healthcare.

2026

Recognition

Named an HFS Hot Tech 2026 company; collects Globee and AI Breakthrough awards.

Round sizes

Where the $85M came from

Seed '23
$6.75M
Series A '23
$20M
Series B '25
$25M

Lead investors: Clear Ventures (seed), Khosla Ventures (Series A), Prosperity7 Ventures (Series B). Bars scaled by round size; total across all rounds is approximately $85M.

If you can explain a process to a coworker, you should be able to automate it - and read exactly what the machine did afterward. - The Kognitos thesis, paraphrased
The obvious questions

FAQ

What does Kognitos actually do?
It lets enterprises automate business processes by describing them in plain English, then executes those instructions deterministically through a neurosymbolic AI engine - with an audit trail you can read back in the same plain English.
What is neurosymbolic AI, and why does it matter here?
It combines the language understanding of LLMs with a symbolic rules engine that runs instructions exactly as written. Kognitos uses the split to interpret intent with the neural half and execute reliably with the symbolic half - the basis of its "hallucination-free" claim.
Who founded Kognitos, and when?
It was founded in 2020 by Binny Gill, former CTO of Nutanix, together with co-founder Nicholas Brigham Adams.
How much has Kognitos raised?
Roughly $85M total: a $6.75M seed (2023), a $20M Series A led by Khosla Ventures (2023), and a $25M Series B led by Prosperity7 Ventures (2025).
How is it different from RPA tools like UiPath?
Instead of brittle screen-scraping bots, Kognitos uses English-as-code automations that business users can build and maintain themselves. The company cites faster deployment, lower maintenance, and high exception auto-resolution versus legacy RPA.

Compiled from public sources including Kognitos press releases, company pages, and third-party coverage. Figures such as employee count, revenue estimates, and vendor performance benchmarks (*marked) are approximate and drawn from company or third-party statements. Where a detail could not be verified, it has been omitted.