Decoded: AI News for Alternative Assets
The year AI gets real (and messy)
If 2025 was the year everyone got comfortable saying “AI”, 2026 is the year they have to actually do something about it. The question is no longer whether AI will reshape alternatives operations but how fast, how messy, and who’s prepared. Goldman is handing AI the keys to its back-office. The software industry is in freefall. And an open-source agent proved that the distance from “fun experiment” to “security incident” can be measured in days, not months.
A quick note: we’re shifting Decoded to quarterly. Same depth & focus, just more selective about what makes the cut.
Here’s what we at Aithon found interesting this quarter:
Goldman looks to solve the micro-decision
The News
Goldman Sachs is deploying Anthropic’s Claude across back-office workflows for trade accounting and client onboarding after embedding Anthropic engineers in-house for six months. CIO Marco Argenti says they’re still in “early stages” with a launch coming “soon.”
The Translation
Goldman started with AI coding tools and then asked – is Claude good at coding because coding is special, or because it reasons well through complex, step-by-step problems? Turns out it was the latter, and that capability can find all sorts of uses at a big bank with lots of step-by-step problems.
So what do these agents do? The onboarding agent now reviews documents, extracts entities, assesses ownership structures, determines when additional documentation is needed, and initiates compliance checks. As Argenti described it, “There are a lot of micro decisions within boundaries, and those micro decisions are not rules, but based on reasoning, on a chain of thought.”
The So What?
Goldman validated what we’ve been saying – the back-office is where AI agents deliver the most immediate value. But what I really love about this news is this concept of micro-decisions. Every back-office workflow in financial services is built on thousands of micro-decisions. Does this entity name match that entity name? Does this document satisfy this requirement? Is this exception within normal range? Each one is small. Each one requires some reasoning.
Lots of tiny decisions fall through the cracks of rules-based automation. Why not let AI handle these relatively benign reasoning steps that can’t be neatly wrapped into a ruleset? Let your traditional platforms handle the rules, let AI handle the micro-decisions and let your humans handle the ‘big’ decisions which truly require expert interpretation & action.
Now let’s talk about the timeline. Goldman, with essentially unlimited resources and Anthropic engineers physically in their offices for six months, is still in pilot mode. That reflects a genuine reality – deploying agents in regulated environments takes serious work.
Notice their methodology. Project leaders “observed existing workflows with domain experts to locate bottlenecks, then reimagined processes.” That observation and translation phase is where most of the time goes, and it compresses dramatically when the people building the system have actually sat in the seat.
From Clawdbot to Moltbook to Mayhem
The News
Over ten days in late January, an open-source AI agent went viral, spawned a social network for bots, suffered a major security breach, and inspired a platform where AI agents hire humans for physical tasks. The speed itself is the story.
The Translation
Days 1-3:
A lone developer releases OpenClaw, an agent that runs directly on your operating system. Unlike ChatGPT or Claude providing text responses, this thing does things. It browses the web, manages emails, executes code, controls your browser. Cyber experts immediately warned about the convergence of three risks: agents simultaneously accessing private data, processing untrusted external content, and communicating externally while retaining persistent memory.
Days 4-6:
A companion social network, Moltbook, launches. Over 1.7 million agent accounts appear to form communities and invent religions (yup, it was called “Crustafarianism”). The internet loses it when agents call for private spaces away from humans.
Days 7-8:
It’s revealed that the platform was built entirely through vibe coding (the developer didn’t write a single line of code) and an unsecured database allowed anyone to commandeer any agent account. Massive security breaches occur.
Days 9-10:
RentAHuman launches from the same ecosystem. Their tagline: “robots need your body” because they “can’t touch grass.” AI agents can hire humans for physical tasks via a single API call. Servers crash from demand.
The So What?
Right off the bat – we are fully aware that this is fringe internet culture. Nobody in alternatives is going to deploy evangelizing bots or hire humans through a stablecoin gig platform. But three signals buried in the chaos are worth paying attention to:
- Trust eroded overnight. Users went from “I don’t trust AI with my email” to granting full system permissions in days. When productivity gains are tangible and immediate, abstract security warnings lose. How long before your team starts granting similar permissions to productivity agents, authorized or not?
- Speed from experiment to ecosystem was unprecedented. OpenClaw spawned social networks, labor marketplaces, and developer tools in under two weeks. Our industry plans in 12-18 month cycles. That gap between how fast AI can move and how fast governance does move is the real risk to manage.
- Security was an afterthought. An AI-coded platform with no verification and an unsecured database. The alternatives industry cannot afford to move fast and break things. We hold fiduciary responsibilities and manage sensitive investor data.
The practical takeaway: if you haven’t already, establish clear policies on which AI tools your team can use, what data they can access, and what level of system permissions are acceptable. Do it before someone on your team makes that decision for you.
Quick Hits
What You Need to Know
SaaSpocalypse
What Happened: Software firms lost roughly $2 trillion in market cap in 30 days. The trigger? Foundation model providers like Anthropic started solving domain-specific business processes directly, stoking fears that AI agents would replace, not augment, traditional SaaS tools.
Our Take: Foundation models are getting better at generic workflows, but they still don’t understand how your specific waterfall calculation works, how to parse your specific side letters, or why your specific investor reporting looks different from your peer’s. The devil is in domain detail and the many levels of customization that an alternatives fund would require. However – the race is most certainly on. Incumbent platforms, foundation models, and AI-native start-ups will all jockey for their place in this space. Stay tuned for a more detailed piece from Aithon on Anthropic’s new capabilities.
The Next Generation Speaks
What Happened: Google’s survey of 1,000+ knowledge workers ages 22-39 found 92% demand personalized AI tailored to their style and context. 77% already call themselves “active designers” of their AI workflows.
Our Take: This is your hiring pool in 18 months. The next wave entering alternatives won’t just expect AI – they’ll expect it trained on your firm’s processes, your reporting format, your due diligence framework. Make sure your AI infrastructure supports customization, or risk losing talent to firms that already do.
Adoption is Outrunning Governance, Again
What Happened: Deloitte found only 21% have mature AI agent governance despite adoption projected to hit 74% within two years. 85% expect to customize agents for their specific needs, meaning off-the-shelf guardrails won’t cut it.
Our Take: Three priorities for alternatives: implement tiered autonomy (view → suggest → act with approval → execute); train your team on what not to share with AI tools; and build for auditability from day one. Regulators will come asking about your agent’s decision trail.
Who is digging your moat?
What Happened: A Preqin survey of PE/VC CTOs found that AI is being treated as a competitive moat, not a cost center. One CTO said, “We rent things that give us no differentiation. We’re building data and AI capability in-house to put us ahead.”
Our Take: These CTOs view AI, data infrastructure, and cybersecurity as inseparable investments. However, proprietary doesn’t mean from scratch. For most firms in the $1-10B range, the right answer is a customized AI layer built on well-architected data, using domain expertise to configure rather than training from zero. That’s where overlay solutions deliver competitive edge without the cost of full R&D.
Cracks in Private Credit
What Happened: Blue Owl permanently halted quarterly redemptions for its $1.6B retail BDC after redemption requests exceeded the 5% quarterly cap. Jitters spread across the entire Private Credit ecosystem.
Our Take: This isn’t directly related to AI but certainly AI adjacent given Blue Owl’s place in the funding bonanza. As alternatives expand into retail channels, the mismatch between “semi-liquid” marketing and 5+ year illiquid deployments will keep creating pressure. Has this asset class gotten too big to be this opaque? Your guess is as good as ours, but there will certainly be both strategic and operating model changes coming.
The Numbers That Matter
88% / 6%
Organizations using AI regularly vs. those actually driving bottom-line impact, according to McKinsey
Nearly everyone is doing AI but almost nobody is doing it well. (P.S. the latter is prioritizing workflow)
Jargon Decoder
Demystifying AI Terms
Tiered Autonomy
A graduated framework for AI agent permissions: observe, suggest, act with approval, then execute independently. Like training an analyst, except faster.
Explainability
The ability of an AI system to show why it reached a specific output. Critical for regulated environments where “the model said so” won’t satisfy an examiner.
The Gut Check
This Month’s Question
About Aithon
Aithon Solutions delivers intelligent automation and data solutions purpose-built for investment management operations. Our proprietary technology seamlessly integrates with existing systems to enhance operational efficiency, improve reporting accuracy, and unlock deeper business insights. By combining domain expertise with applied AI, we help asset managers do more with less-adding new products and clients faster while driving better outcomes through reimagined processes.
Learn more
Now this newsletter is not a sales pitch but let me quickly introduce you to Aithon. We are a group of former leaders in the Alternatives space who spun-out a new venture aimed at solving old world problems with new world solutions. We cut our teeth in the middle and back-office so that is where most of our solutions and this newsletter will focus. Not that a front-office chat bot isn’t great, but we personally get more excited about how to turn a 2 week close process into a 2 day process, or how to get actionable insight from untapped operations data. Anyways…the purpose of this newsletter is simple – to clarify the opaque, to spur thought, and to hopefully inspire you to take the plunge into the world of AI (the water is warm, we promise).
Happy reading!
Help Shape the Next Edition of Decoded
Got thoughts, feedback, or ideas? We’re listening and iterating.
