Decoded-AI-News-for-Alternative-Assets-Sep-2025-min

Decoded: AI News for Alternative Assets (Sep-2025)

by
with no comment
Newsletter

September 2025 Newsletter

Decoded: AI News for Alternative Assets

Beyond Silicon, Into Substance

The Al news ecosystem is a funny, cyclical place. Big number > bigger headline > more engagement > more resharing > wait for an even bigger number > repeat. Specifically, this month was dominated by news of bigger gigawatts, bigger compute deals, bigger capex and bigger valuations. Those who read between the lines were rewarded with some smaller but equally important developments towards actually operationalizing Al.

Here's What We Found Interesting This Month:

  • Agentic breakthroughs make the "why not?" question impossible to ignore
  • The great 'retailization' of Alts is forcing a fresh look at ops transformation
  • Autonomous coding should redefine "juice not worth the squeeze"
  • Hallucinations are a feature, not a bug. Manage risk accordingly
  • Finance leaders are skeptical, but measured pilots can help drive buy-in
  • Shiny Al output is often busywork in disguise
  • Big tech pays big to train agents in complex, real-world environments
  • Private markets CTOs are treating AI as a competitive moat

Google Lets Users Take Their Hands Off The Wheel

The News

Google made their first major step into agentic browsing by adding a Gemini assistant directly within their Chrome browser. Two days earlier, they launched a new Agent Payments Protocol (AP2) that lets Al agents make secure purchases on behalf of individuals

The Translation

The new Chrome release allows users to ask for information or analysis about a webpage without opening Gemini separately (or another LLM...). This first release also offers basic integration with native Google apps like YouTube, Maps, etc. The big deal here is their plan for a true agentic browsing assistant. This means Al will be able to take action and execute multi-step tasks ranging from ordering groceries to booking a vacation


Separately, the AP2 protocol enables Al agents to prove they have permission to make payments, ensures requests reflect user intent, and builds in accountability through traceable audit trails. The protocol has major backing from very large & reputable companies in the payments & banking space.

The So What?

What I personally love about Al is it is now forcing us to ask questions about how we go about our lives and how we go about our work. Nearly every day I ask myself why do I do X this way? Why not another way? Let's explore this chain of thought:

  • Why do market research by scanning endless websites and white papers? Why not use an LLM?
  • Why type prompts in an isolated LLM? Why not an agentic browser like Comet?
  • Why use a separate agentic browser? Why not Chrome since it already knows my browsing history?
  • Why analyze and aggregate this information into PowerPoint? Why not let Al do it directly?

I like to think that we are living in the era of efficiency, and with each iteration of Al and each reflection of "why", we come one step closer to STP'ing our lives.

 

With that said, why not.. Al-executed payments? Let's pretend I receive a legal invoice every month. They are fairly consistent but I do a sense check before instructing my team to initiate the wire in a separate payments platform. This is how it would work in an agentic framework leveraging AP2:

  • I delegate legal fees to mv navments agent with criteria on timing, amount thresholds, and any other conditions. This Wall street meets main street (with Al as the; akin to standing instructions in AP2 terms.
  • My data extraction agent pulls payment requests directly from PDFs in Outlook and directs them to the appropriate sub-agent.
  • My payments agent confirms the payment request is within the bounds set by the intent mandate.
  • It then creates a 'cart mandate' on my behalf allowing for autonomous payment.
  • The complete sequence creates an audit trail with full accountability

I'm intentionally oversimplifying here... but recognize that we are going through a fundamental shift from Al systems that recommend actions to systems that can execute autonomously. Why not join in?

Wall Street Meets Main Street (With Al As The Matchmaker)

The News

On the heels of relaxed U.S. policy, Goldman Sachs & T.Rowe Price tie up to draw retail wealth and retirement savings into alternative funds. In separate but related news, a new crop of technology providers have emerged to support this "retailization" of alts.

The Translation

Goldman and T. Rowe are just one example of the alts to traditional fund partnerships which seem to be in vogue for 2025. We also have Legal & General + Blackstone, BlackRock + Partners Group, Vanguard + Wellington + Blackstone, amongst others..

 

These partnerships share similar threads new wrappers for old assets, high-scale distribution, liquidity engineering, and risk optics. Oh and fees, let's not forget about the fees.

 

To support these new marriages, new Al-powered firms are building the infrastructure and front-end gateways where funds and their fun-sized investors will come to do business. Solutions range from curated product selection & distribution, data extraction & aggregation, KYC / AML and real-time reporting.

The So What?

This is a significant ecosystem shift forces managers to transition from opaque models optimized for a small pool of institutional investors to a high-scale, high-transparency model suited to us normal folk. More accounts, more transactions and more exceptions will stress all parts of the operating model:

  • To what extent will the AML/KYC onboarding process need to be strengthened?
  • How will fund structures evolve to accommodate new liquidity mechanisms?
  • Will we need more frequent or transparent portfolio company valuations?
  • What can we expect in terms of fee structures, disclosures and calculation methodologies?
  • How deep and how 'real-time' must reporting go?

These are questions we may not have immediate answers for but we can make some pretty safe assumptions about what they imply in terms of technology:

  • Workflows need to be automated as there will no longer be room for manual, offline processes
  • Data must be centralized, structured, and reliably governed for both operational efficiency and audit-readiness
  • The document conundrum must be solved both going into (e.g. KYC docs) and out from (e.g. capital calls) the fund manager's ecosystem
  • Demands for real-time GenAl chat interfaces will mimic the demands seen in the consumer market

There will be competitive opportunities and pressures on both the top- and bottom-line. Those who not only ‘adapt’ but truly transform their operating model to be data-first, Al-enhanced will be best placed to win.

Quick Hits

What You Need to Know

Replit Achieves "Full Self-Driving Moment" For Software Development

What Happened: Agents can not only code, but can autonomously handle complete dev workflows from ideation to deployment

Our Take: We don't use Replit directly (this is a sponsor-free newsletter) but it's clear that custom tool development just got a whole lot more economically viable. Rapid prototyping, testing and deployment should prompt you to look at workflows you previously deemed "juice not worth the squeeze". If your roadmap and time-to-ship looks the same in 2025 as 2024, you're probably doing something wrong...

Open Al's Honesty Moment

What Happened: OpenAl admits hallucinations will never completely disappear.

Our Take: Hallucinations were long ago (like a month ago in Al-scale) thought of as a solvable bug which will go away with more training. Now it's clear that eliminating them completely is fundamentally unavoidable (as is 100% accuracy). Risk management protocols and human-in-loop need to be pillars of any Al use case, at least until LLMs stop incentivizing guessing in training.

CFO Remain Ever Skeptical Of Al.... Rightfully So?

What Happened: CFO surveys show a similar distrust in Agentic Al as shown with Generative Al two years ago.

Our Take: Lack of control, unclear ROI and unreliability are totally valid concerns…but watching from the sidelines is not the right answer. In practice, implementing AI is more like an operational transformation than it is a technology project. And like any large transformation, baby steps will are key to bringing your team on the journey. Pick something safe, something measurable, and something that will start building trust.

Is Al "Workslop" Destroying Your Productivity?

What Happened: Low-value Al generated work looks great but is often a pointless drag on productivity

Our Take: With a shiny new kit of productivity tools in hand, knowledge workers are skipping the knowledge and putting out work for the sake of...work? Why give my colleague a concise, well-thought 200 word analysis when I can ask Al to spit out a beautiful, 15-page research paper in 2 minutes? HBR put the solution more eloquently than I could "indiscriminate imperatives yield indiscriminate usage". It's your job as a leader to both set the rules and the tone for Al usage.here to stay.

The Billion Dollar Al Gym Membership

What Happened: Major Al players are investing heavily in reinforcement learning environments where agents can learn complex, multi-step tasks under real-world conditions.

Our Take: As our Al agents grow stronger, they will eventually need a gym upgrade. Think of my legal fee example earlier and how much more confident you would be if those agents were to do job training on a full calendar year of invoices and accounting entries in their source systems. Sounds good, right? Unfortunately, this new training methodology is heavy, expensive and currently limited to big tech. Expect costs to come down to a point where this would be a more feasible option for fund managers.

The Numbers That Matter

72%

Portion of Al models in production that fail within 12 months due to poor data pipelines Garbage in, garbage out. The best Al in the world can't fix bad data architecture.

Jargon Decoder

Demystifying AI Terms

Vector Embeddings

What it means: Representing text, images, or data as numerical coordinates so Al can "search by meaning". Like mapping every deal term sheet into a single multidimensional database of comparables.

 
 

Prompt Injection

What it means: When someone tricks your Al by hiding malicious instructions in seemingly normal input. The Al equivalent of social engineering attacks.

 
 

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.

Know more
Tags: No tags

Comments are closed.