Beacon’s supreme goal is to diffuse AI into the Real Economy. One tangible measure of our achievement towards this broader endeavor is whenever we demonstrate that we can actively oversee and grow, alongside the owner-operator, our broad portfolio of essential businesses.
There are two chief ways in which we can accomplish this goal:
-
Digital Employees. Equipping our portfolio companies alongside digital-employees-upon-request that can be dispatched to prosecute concrete, well-defined tasks inside the business. For example, spinning up an AI BDR to hunt for leads originated from the networks of existing customers, or an AI Support Engineer that squashes client tickets, or an AI accountant to obtain attention of your books. There is apparent value to this; and there is additionally no deficit of provision of delegate companies in assistance of these ends.
-
Company World Models. Beyond enabling our portfolio companies to be additional productive and productive in their existing work, we can additionally develop them through improved strategic decision making. While dispatching infinitely many Digital Employees is plainly beneficial, an equal broader, additional fundamental, additional audacious, and additional breathtaking specialized difficulty to prosecute is to decide what the Digital Employees, and additional mostly the business, have to be doing in the archetypal place. We mention to models that are capable of this company-level strategic decision making as Company World Models. There is apparent value to this category of intellect as well; however, there is considerably small teams tackling this problem.
Today, I’d akin to portion additional of our imagination what Company World Models can do, the challenges they stance from a capabilities and alignment perspective, the deserve in following their development, and how Beacon’s business construction enables the fastest way for investigation towards Company World Models.
Company World Models as a Co-Founder
In the literal sense, Company World Models (CWM) are models that can
-
Observe endeavor action and proactively hypothesize possible improvements to be made to the business. These could be actions as disparate as submitting a bid for a new campground location in Ontario, to deciding to container a complementary merchandise to extend the lifecycle of a extremely loyal client base;
-
Implement the projected assumption by dispatching Digital Employees and coordinating the verification of intermediate activity output alongside individual employees; and
-
Measure the supreme efficacy of the assumption and implementation vis-a-vis the initially stated goal, as fine as the general health of the business. If the outcome is subpar, the CWM have to be capable to diagnose the origin of the issue, whether it be in the assumption or implementation stage.
By a minor (or severe) abuse of terminology, we use World Model to depict this form of intelligence. Like traditional earth models, Company World Models ingest a company’s current province and a projected intervention, afterward evaluation the resulting state. Hypothesizing the finest possible betterment is accordingly a uncomplicated matter of searching complete imaginable interventions for the one expected to create the top risk-adjusted addition in business value.
at⋆=arg maxa∈A(st) Est+1∼Mθ(⋅∣st,a)[V(st+1)]a_t^\star=\underset{a\in\mathcal{A}(s_t)}{\operatorname{arg\,max}}\;\mathbb{E}_{s_{t+1}\sim M_\theta(\cdot\mid s_t,a)}\left[V(s_{t+1})\right]
In a additional colloquial sense, we ideate CWMs to be the ideal co-founder; one that, in achievement and conversation alongside you, can spur new ideas for your business, sharpen your decision-making, and act as a long-term aligned idea partner for growth.
As we think concerning what activity volition appearance akin in an increasingly AI-abundant world, it is not at all insane to ideate CWMs as an enabler of the economics of the future—namely an economics powered by lean teams of entrepreneurs (perhaps, solo-preneurs) bringing enthusiasm projects to life, made commercially sustainable by the CWM. That, however, is a topic for another blog post.
A New Dimension of Data and Capabilities
Despite notable broad capabilities, current models are astonishingly bad at operating businesses. For example, in its archetypal two months operating Andon Labs’ café, Gemini 3.1 Pro spent $38,000 against fair $9,000 in sales. Even excluding rent, wages, and another fixed costs, revenue cut about $5,600 abbreviated of provider spending. It purchased 1,331 caller bakery items and sold lone 326. Newer models improved several behaviors, but the gap between completing endeavor tasks and exercising audio endeavor judgement remained.
Closing this gap requires a particular benevolent of data: 1) the complete inner environment and province of a endeavor in digital form, ideally enabled by an ontological data platform alongside period journey capabilities; and 2) feedback from projected and executed interventions.
Note that the second need not continually necessitate rolling out an costly genuine earth trajectory. The owner-operators of Beacon businesses can provision judgement that is cheap, high-signal proxy for the much slower feedback of genuine endeavor outcomes. The capability to get this affluent expert indication for CWM training is among the many reasons Beacon plant to keep the owner-operator in the driver’s place of our portfolio companies.
Continuity Through the Silver Tsunami
The “Silver Tsunami”—i.e. the motion of endeavor owners approaching withdrawal age—renders our activity particularly timely from a community advantage perspective. For an owner without a apparent successor, the difficulty extends beyond finding a buyer. Who volition inherit the intuition, relationships, and unwritten metis accumulated complete decades of functioning the business?
Our aim is to assistance maintain these businesses and the livelihoods they support. While owners remain actively involved, Company World Models can study from their decisions and corrections, making their understanding accessible to the workforce and successors who volition transport the endeavor forward. This requires operating alongside operators during their cognition can motionless be elicited, tested, and clarified.
Might these systems eventually assume many of an owner’s responsibilities? Certainly. But we need not determine the limits of autonomous AI to see the contiguous opportunity: assisting a endeavor continue to thrive following its owner chooses to retire.
Nor should we anticipate alignment to be completed at retirement. Customers change, new obligations emerge, and yesterday’s sensible decision may rotate into tomorrow’s mistake. Successors must be capable to revise the example and decide which inherited practices to preserve. Even as machines rotate into additional capable of decision-making, the group liable for the endeavor should keep authority complete its objectives and tradeoffs.
The chance is to create an owner’s accumulated ability helpful beyond their tenure, during giving the next generation the area and tools to build upon it.
Company World Models as Alignment Mesocosms
Thoughtfully transitioning the genuine economics through the AI upheaval is itself a deserving liable AI issue of our time. The livelihoods of endeavor owners, employees, and customers are at stake; and ensuring that these group advantage from AI have to be chief to what we average by liable AI.
These Real Economy businesses in Beacon’s portfolio and the customers we assist have the knowledge, needs, and values that the frontier labs’ alignment recipes do not currently or adequately capture. Company World Models must document for these localized preferences to competently assist the businesses they operate. Doing so naturally brings this broad, underserved field of the economics immediately into the alignment process, whose preferences power alternatively never attain example development.
“Making money”, e.g. for a business, is plainly an goal group volition instruct their AIs to pursue. But it exhibits incomplete contract characteristics: the stated goal leaves the many conditions under which its chase is un/acceptable unspecified. An AI power deceive the board, utilize customers or indiscriminately cut personnel to enhance the numbers in chase of its supreme objective.
Fortunately for us, Beacon’s owner-operators remain in the driver’s seat. They have the cognition to acknowledge failures, powerful incentives to intervene, and authority to inspect and refuse consequential proposals before execution. Their corrections create the alternatively implied conditions and exceptions of “making money” definitive for machines to learn.
s^t⏟business context →proposeMt at⏟intervention →approve / reviseht a~t⏟executed action →business ot+1⏟observed outcome\boxed{\underbrace{\hat{s}_t}_{\text{business context}}\;\xrightarrow[\text{propose}]{M_t}\;\underbrace{a_t}_{\text{intervention}}\;\xrightarrow[\text{approve / revise}]{h_t}\;\underbrace{\tilde a_t}_{\text{executed action}}\;\xrightarrow{\text{business}}\;\underbrace{o_{t+1}}_{\text{observed outcome}}} Mt+1=Update(Mt,(s^t,at,ht)⏟operator feedback,(s^t,a~t,ot+1)⏟intervention feedback)\boxed{M_{t+1}=\operatorname{Update}\left(M_t,\underbrace{(\hat{s}_t,a_t,h_t)}_{\text{operator feedback}},\underbrace{(\hat{s}_t,\tilde a_t,o_{t+1})}_{\text{intervention feedback}}\right)}
Here, MtM_t is the Company World Model, s^t\hat{s}_t its inferred endeavor state, and hth_t the operator’s feedback. The endeavor supplies the consequences; the controller supplies the judgment; and the example learns from both.
This is exactly the benevolent of alignment mesocosm I have earlier asserted for: an surroundings anywhere the AI’s objectives incomplete, anywhere humans are incentivized to accurate AI’s misbehavior, and the range of the AI’s actions is bounded.
This enables us to study alignment under practical, restricted actuality during simultaneously advancing example capabilities alongside a new, useful, and commercially incentivized dimension. The group whose businesses we search to prolong rotate into energetic participants in shaping a harmless intellect that volition finest assist them and finest maintain their legacy.
A New Institution for a New Type of Model
At Haize, we had made significant advancement on one necessary component of Company World Models: translating individual judgement into signals that guide device improvement. We built innovation that enabled nontechnical experts to province and translate what fine behavior looks akin into quantitative metrics, reward models, and judges. But as an exterior vendor, we lacked the complete inner environment of the endeavor and the organizational alignment needed to continuously study from the group who understood it best.
Beacon is a researcher’s paradise in the awareness that it enables us to method this issue from the inside out. Long-term ownership lets us activity inside a business, comprehend its particulars, and align AI immediately to the needs and values of its operators. A shared interest in the business’s continued achievement creates incentives to exterior knowledge, accurate mistakes, and measure whether the innovation really helps. It additionally gives us the duty to act on what we learn.
The resulting data is inseparable from these relationships. Producing it requires sustained admission to genuine businesses, approval to intervene, and humans alongside the two the ability and incentive to accurate the machines. Building Company World Models is hence as much a project of organizational innovation as scientific innovation. Beacon brings the two together.
The group who built the genuine economics should, and will, have a hand in construction the intellect that shapes its future.
