Half of all enterprises are now shifting away from renting artificial intelligence through closed APIs, choosing instead to bring open-source models in-house[1]. They are tired of building their operations on rented land. If you run a wealth management firm, you might assume this infrastructure debate belongs to software companies. It does not. The tools you use to draft client emails, summarise meeting notes, and speed up paraplanning are fundamentally changing.

We have spent the last three years in the era of the chatbot. We are now entering the era of the agent. This means the system is no longer just answering questions. It is executing multi-step workflows across your business applications.

The shift to autonomous execution

The latest model releases treat artificial intelligence as an autonomous decision-maker rather than a passive tool. Instead of waiting for a human to prompt every step, an agentic system takes a goal and works through the required actions to complete it across different pieces of software.

While 75% of businesses express interest in these agentic workflows, only between 11 and 17% have actually deployed them[2]. Even more tellingly, 40% of agentic projects are cancelled due to deployment challenges[2].

The gap between ambition and execution is vast. Connecting an AI to your CRM, your document management system, and your email requires a fundamentally different architecture than simply buying chat licences for the team.

You are granting a system the permissions to take action on your behalf. That introduces security and compliance considerations that a standalone web chat simply does not have.

Why firms are moving away from rented models

The economics and the risk profile are pushing firms toward smaller, self-hosted models. For the past few years, the standard approach was to send your data to OpenAI or Anthropic. But relying entirely on closed APIs introduces a severe platform risk.

Anthropic recently reduced the cost of its Claude Sonnet 5 model[3]. Despite these price drops for cloud models, firms are increasingly prioritising control over pure capability. The lack of control over data processing and the risk of sudden access changes make closed APIs a liability for core infrastructure. Firms are now favouring narrow, open-source models that cost less to run and operate entirely within their own secure environments.

Securing your workflows

If you are integrating artificial intelligence into your firm this year, you need to rethink your permissions. Running any agent with broad access creates a massive security surface area.

First, audit your API dependencies. Look at the tools your firm relies on for daily operations. If your entire workflow breaks because a vendor raises their pricing by a multiple of three or changes their terms of service, you have a platform risk you need to mitigate.

Second, restrict agent permissions. An AI agent should never have global read and write access to your client database. Compartmentalise access so that a hallucinating or compromised model cannot overwrite critical compliance records or client files.

Third, keep the human in the loop. Autonomous execution does not mean unmonitored execution. A system can draft a suitability report and route it for internal approval, but a qualified human must always review and sign off before it reaches a client or goes on file.

Matching the tool to the problem

Most deployment failures happen because firms buy complex platforms to solve simple problems. You do not need to build custom infrastructure if an existing tool already does the job.

If your advisers just need help summarising meeting notes, you are looking at Level 1 (education). An off-the-shelf tool that supports security, with the right prompting, will solve the problem today.

If you want those notes to automatically update the client record in your CRM and draft a follow-up email, you are looking at Level 2 (integration). You connect two or three tools together using software like Zapier or Make.

Only when you need a highly specific, multi-agent workflow that interrogates proprietary datasets do you move to Level 3 (custom build). The best defence against project failure is to start at the lowest level that genuinely solves the problem.

The shift from rented chatbots to owned agents is going to create a divide between firms that build secure, integrated workflows and those that just pay for more software subscriptions. If this is the situation your firm is in, a discovery call with Cordrey Consulting is a good place to start.

This article is for informational purposes only and does not constitute regulated financial advice or a compliance opinion. Consult a qualified compliance professional for advice specific to your firm.

Sources

[1] TechCrunch, ‘Hugging Face’s CEO on why companies are done renting their AI’, TechCrunch, 10 July 2026. Available at: https://techcrunch.com/2026/07/10/hugging-faces-ceo-on-why-companies-are-done-renting-their-ai/ [2] Digital Applied, ‘Why Agentic AI Projects Get Canceled (and How to Ship)’, Digital Applied, 11 July 2026. Available at: https://www.digitalapplied.com/blog/agentic-ai-project-cancellations-gartner-40-percent-2026 [3] Anthropic, ‘Claude 3.5 Sonnet’, Anthropic Announcements, 1 July 2026. Available at: https://www.anthropic.com/news/claude-sonnet-5