The next billion software users will be agentic systems.

General Agentic is bringing enterprise software to the next billion users, enabled by our vertically integrated software supply chain.

Scroll
Purpose
1 billion

global activeenterprise agentsby 2030

40x morethan in 20251

$0

of workgets doneby agents

that cannot accessthe software2

18%

of existingenterprise APIspublicly accessible

locking outagentic users3

Vision

Most software usage will arrive through agentic systems.

Not humans clicking buttons on screens.

Most acquirers see AI as a cost story. They run what they buy with less, often rebuilding working systems from the ground up, a gamble their customers cannot afford.

We see a growth story. We unlock software already in production for its next generation of users, repackaged for agents, without disrupting the pathways today’s users rely on. We believe this opens a far larger market at a minimal added cost.

Systems

One system, from the network layer through the business logic.

A reference implementation of the complete software supply chain, built and operated by General Agentic. Multi-tenant, cloud-agnostic, and model-independent from the ground up.

Network

Our own network, compute and data fabrics lead to fixed-cost compute. Owning the root of the stack keeps the economics predictable as agentic usage grows.

Supply chain

The machinery that moves software to its new users: continuous integration with real-time delivery, identity, connection APIs, permissioning and governance, and an SDK for seamless handoff. One integration at the core, not one per platform.

Business logic

The validated logic of the software we acquire, served to agentic systems at full fidelity. We do not rebuild it, we deliver it.

Our founders run the firm on this system, including the application that built this very page. We were its first users and the companies we partner with are its next.

Partnership

We partner with companies through acquisition.

Our process

We acquire established software companies and own them for the long term. Your name stays on the door, your customers keep the product they rely on, and when their agents arrive, it is still your program powering their work.

Your customers bring their agents, you bring the logic they trust, and we bring the technology that connects them.

In this era, software splits into a rewrite or a repackage. We buy where the natural path is a repackage. We keep the system of record, we do not rebuild the core, and we connect your product, as it exists, to every agentic workspace your customers use.

Phase 1 (First and Second Quarters)

Connect identity, leave the product alone

Nothing changes for your current customers. We connect your identity system and APIs to a governed distribution layer, and together we configure permissioning so an agent can work at the same fidelity a human user can. You control what is exposed, and to whom. People keep the sign-in they already have.

Phase 2 (Third and Fourth Quarters)

Open a governed route into the workspaces they already use

Your product becomes reachable from ChatGPT, Claude, and the other surfaces where work is starting. To users, it's the same product, under the same entitlements, just delivered through a new channel. Usage starts to climb, and we can test pricing models that are a win-win with the customers rather than only seats.

Phase 3 (Second Year Onward)

Ship the roadmap through that route

Development continues as you planned. Every new feature can reach those workspaces without a second architecture. Your team inherits the tooling to do that at product velocity. Your customers bring their own agents and pay their own inference. The roadmap starts growing where the new users are.

The results show up where it counts: new users on the same product, more revenue per customer, and a channel you can price, because the work now starts where your customers already work.

How we help

Growth is our focus, efficiency is our engine. We study the frontier of Silicon Valley and Wall Street, prove it out on ourselves, and install what works in the companies we own. We relentlessly source and test tactics and technologies that help our companies grow faster, more efficiently.

Growth

Build
Product Development & Velocity

We help you build for agents, faster. This requires new surfaces, new permissions, and new features designed for agentic users. Your team starts from our working system, not a blank page.

Sell
Sales, GTM & Customer Success

New users deserve new pricing. Agents don’t buy seats, so we help you price what the product actually delivers: usage, tiers, and outcomes without breaking contracts. Buyers increasingly choose software their agents can reliably use, and every agent they deploy expands the account while tying your product deeper into their work.

Efficiency

The stack
Software Supply Chain

Success in the agentic channel multiplies usage, and metered infrastructure bills you for it. Over time the supply chain centralizes onto the one system we operate, the way a well-run fleet is renewed: opportunistically, piece by piece, as each part reaches the end of its useful life.

The business
Finance, Accounting & Back Office

Every company carries work that has nothing to do with its product. Finance, accounting, payroll, and reporting are centralized for every company we own. Your team’s time goes back to making the best products for your customers.

What we don’t change

You spent years earning your brand and your customers’ trust. That is the asset we are buying. There is no fund clock and no exit deadline, since we aim to create value for shareholders by growing operating earnings from a group of controlled businesses over a long time horizon. Our technology and our capital exist to compound what you built through the next platform shift.

Company

General Agentic was founded on a straightforward principle:The best acquirers of technology companies are builders of technology themselves. The firm pairs Silicon Valley technical discipline with Wall Street investment judgment, and operates from New York.

The firm’s founders bring experience from world-leading institutions.

goldmansachsstanforddeshawwhartonciscosystemsdukesapphire
Thoughts
01 / 03
Thought Piece8 min read

The Missing Metric

AI is missing its “miles-per-gallon” equivalent. Token counts and ambiguous model SKUs are not the answer. Until there is an obvious proxy for outcomes, value-based pricing for AI products will be unclear.

By and · September 2026 · New York, NY

You either own, you loan, or you pay-as-you-go. That’s a perpetual license, a subscription, or a consumption-based pricing model.

In any market, supply is priced at a premium to cost, and demand is priced at a discount to value. Cost is straightforward, but value remains unknown in the AI era.

Let’s think about the combustion engines that powered the planes, trains, and automobiles. All of them from a value perspective could be boiled down to one metric: miles-per-gallon (MPG).

Now think about today in AI. Try to value the intelligence engines that power workflows. Try to think of an equivalent to miles-per-gallon and you’ll run out of human and digital tokens before you reach an answer.

Let’s break down what miles-per-gallon is: the numerator is a unit of distance, and the denominator is a unit of measure for fuel that got you that unit of distance. Given the distance you need to travel and the MPG rating, you know how much gas to put in the tank given your risk level and budget constraints to reach your destination.

Applying this same logic to the cloud era yields another straightforward result: hard drives are measured in bits-per-drive. The numerator is a unit of how many photos can be stored; the denominator is a discrete physical box.

For AI, the market view is “tokens” of all types: input, output, cached. Tokens are not a metric that business users can rely on because they have no tangible meaning and they are also non-fungible. People inherently know that a photo stored at 200kbs is a worse resolution than the same photo stored at 2mbs. People also inherently know that Chevron 91 has better miles-per-gallon than Valero 87. Customers have no way of knowing whether “2 billion input tokens used by 20 headless agents” is better than “200 million output tokens used by 5 sessions of an AI app.”

This brings us to what we call “the missing metric.” The market today has not decided what the “miles-per-gallon” should be for the AI era. That is why pricing remains so opaque and difficult to understand.

Zach and I demystified the blockers to the market finding this metric by going back to a process we learned at Goldman Sachs. When a client would come in wanting to fund a merger but unsure of the direction they should go in, we would show them what we called “the menu page,” which consisted of equity on one side and debt on the other, with every shade of security in between.

We replicated this for the spectrum of pricing options that a business could have here:

The Menu of Pricing Options

01 / 07
Perpetual license + serviceSubscription by seatsSubscription by seats + usageConsumption / usageWorkflowOutcomeTake rate
Revenue behavior
One-time sale, then maintenance
No expansion with use or agents
Recurring per distinct user
Grows with user headcount, not with work or spend
Base per distinct user plus overage when usage exceeds limits
Grows with headcount, and with work
Meters usage, no ceiling other than getting cut off
Grows with thinking, not with dollars made
Paid when a workflow is run
Regardless of whether the workflow generates end value or just costs money
Paid when a defined result happens
Reduced or no cost if defined result does not happen
Tax on a metric associated with a deliverable or a sale
The more ingrained the workflow, the higher the cost
Impact of AI Costs
More support tickets given new tech
End customer pays upgrade costs which puts added friction on agentic rollouts
Cost of rollout is minimal given SaaS
Inference and talent scale with use, making the seller absorb the cost
Seller absorbs only up to a pre-agreed cost
Overage is passed through to buyer
Inference billed through to buyer on the same meter as COGS
Costs felt in real-time, not absorbed
More capable AI means more complex workflows run in AI
Lower setup and maintenance costs
Borne mostly on the seller as the buyer only pays if the result is delivered
Incentivizes better product quality
Passed through directly to the buyer as a % of a metric
The more spend, the more revenue the seller makes
Customer pros / cons
Pro: customer cannot lose an integration or a workflow overnight
Con: updates slow and systems lag the market in capability
Pro: labor-like costs (AI inference) are being absorbed by the seller
Con: seat prices a human not work, so agent usage unbilled
Pro: buyer able to test until value is seen at no new cost
Con: buyer absorbs costs if usage skyrockets
Pro: transparent pricing, cancel anytime to floor cost
Con: unpredictable commodity pricing
Pro: price the job already bought today
Con: likely not a perfect fit to needs if non-custom pricing
Pro: only pay for successful outcome
Con: each payment bundles in the cost of each loss which can make price very high
Pro: pay scales with increased GMV
Con: tax is direct drag on bottom line profits as one grows
Vendor pros / cons
Pro: customer pays more for bespoke service and support
Con: hard to get to economies of scale with many versions
Pro: customer sticky and hard to rip out
Con: vendor absorbs all variable costs, and customer games the seat count
Pro: customer sticky given value accrued by included usage
Con: still eating some of the usage although capped
Pro: profit margins are fixed to vendor
Con: customers are less sticky and have incentive to churn
Pro: setup costs amortized across multiple customers
Con: constant re-working
Pro: quality is proven with each deliverable each time
Con: cost of each failure is variable on the vendor
Pro: fixed take, no COGS surprise
Con: harmed by underlying downturn in client’s business
Requirements
Systems that can be run in isolation
Teams that can manage and upgrade multiple versions
Software that is user-based; in the cloud
Software that can be multi-user and multi-tenant in cloud
Separation of the distinct users from their distinct usage
Seamless detachment of fixed & variable
Discretization of offering into chunks with specific value
Real-time pricing and costs of supply chain
Separate proprietary / unique elements
Workflow creation and run-rate costs well-defined
Judgement of the final product’s quality
Understanding seller’s position in value chain
Clearly defined metrics that can be taxed
Seller who can cap costs into fixed rates
Examples
Oracle: on-prem database engine; capex + support
SAP: on-prem ERP + support
Salesforce: named CRM users
Slack: active users
Workday: priced on people, not payroll
DocuSign: seats + envelope quota
HubSpot: seats + marketing-contact allowance
Anthropic API: meter tokens; pass through inference
AWS: meter data center resource usage + marketplace
Zapier: per successful task
Checkr: per background screen
Avalara: per tax calculation
Zendesk AI: $ per successful resolution, not per attempt
Salesforce Help AI: $0 paid if customer asks for human
Stripe: take rate % + fixed $ per charge
Shopify: card take rate + fine for off-platform transactions

The internal situation that led to this was the July 2026 release of three new model classes by SpaceX (Small, Grok 4.5), OpenAI (Medium, ChatGPT 5.6 Sol), and Anthropic (Large, Fable 5). The pricing was what is known as a Fermi difference, meaning that Fable was 10x more expensive than Sol, which in turn was 10x more expensive than Grok. The industry says “measure cost in tokens.” That’s analogous to measuring gas in octanes. Chevron 91 is not the same as Shell 91. What does octane 91 even mean? Octane 89? How about 87?

To get to that point we need to align on where we are in this technological cycle:

The Phases of the Enterprise AI Platform Shift

02 / 05
Phase 1 ThinkingNowPhase 2 ActionPhase 3 HeadlessPhase 4 AutonomyPhase 5 Value
Breakthrough
Chatting
Tool Calling
Headless Agents
Agentic Commerce
Opportunity Cost
Meaning
A computer program can “think”, and its responses are not constrained by pre-programmed logic.
“Thinking” becomes action, with the models reading from, writing to, and integrating with the systems of record.
Agent runs in the background with pre-configured tools not tied to an AI Workspace (e.g., Cursor or Claude).
Agent makes buying decisions over the tools it can access to execute with a budget given by a principal user.
Agent can set a price for its services and charge counterparts for the value accrued from performing work.
Interaction Mode
Human users install app and message with AI.
Human users select and configure the tools.
Human assigns work, agent comes back.
Human sets budget for task, agent executes.
Services assigned at market clearing price.
Consumer Friction
Need computer access.
Find tasks requiring tools.
Finding many tasks.
Setting up a wallet.
Value proposition itself.
Enterprise Friction
Identifying why chatting with an app is valuable
Getting permission to enter data into system.
Identifying workloads which can show value
Procuring access to an enterprise system itself.
Accountability of who is responsible for issues
Budget cycles of how much can be spent.
Rewiring of the financial rails for agentic spend
Organizational structure of agents as co-workers.
Switching costs are daunting and distracting
New system is not 1:1 to legacy trusted solution.
B2C
2022–2025
2025–2026
2026–present
Late 2026
—
B2B
2024–2026
2026–present
—
—
—

By the end of Phase 3, agents will be headless and we will start to see a metric analogous to miles-per-gallon for agents, and by Phase 4 we will be able to estimate this right after filling up the tank. By Phase 5, we will have enough information to estimate this value given the price ahead of even getting to the pump, and that is where the agentic economy starts to have an opportunity cost.

Jake Chasan & Zach Evans

Co-Founders of General Agentic