Quit tinkering with DIY GenAI pilots: Salesforce’s Agentforce showcases the future of software solutions

After two years of CIOs and IT services teams’ tinkering in their garages to hack together 1,000s of generative AI (GenAI) pilots using Microsoft Copilot, OpenAI’s ChatGPT, and an ever-growing number of large language models (LLMs), one thing is clear: the DIY approach to building your GenAI homegrown solution is slow, risky, and expensive. As CEOs rush to put wins on the board, GenAI solutions forced their teams to cobble together immature solutions and ask every service provider to pitch proofs-of-concept (POCs) to eager business leaders.

The harsh truth is organizations have a target-rich environment for major productivity improvements because existing business processes are built on software applications that stalled their ability to produce significant business productivity and efficiency improvements after the initial implementation. Quarterly SaaS feature releases, enhanced integration capabilities, and worthless Hungarian language support-like additions have not moved the needle for companies. So, hacking together homebrew solutions was companies’ only choice when GenAI debuted. This is changing with the release of agentic AI capabilities by major software providers that deliver crucial new capabilities.

Salesforce’s introduction of Agentforce yanks IT tinkerers out of the garage

Salesforce’s big announcement at Dreamforce 2024 was the introduction of a native GenAI agent development platform called Agentforce. Salesforce promises its Agentforce offers fully integrated capabilities across its SaaS offerings.

Agentforce is coupled with Agent Builder, giving business leaders unfettered access to developing personalized GenAI-powered solutions for their firms without the assistance of IT departments. The solution includes a trust layer, guardrails, toxicity detection, and auditable traceability. CEO Marc Benioff pitched that the Atlas 2 engine, based on the company’s comparisons with current Microsoft and OpenAI solutions, will produce 90–95% project resolution with 33% better accuracy, double the relevancy of outcomes, and produce significantly lower hallucination rates.

If true, sales, marketing, service, and order teams using Salesforce’s solutions should simply dump their DIY garage-grown solutions because the offering is more effective and nearly free of IT’s lethargic pace, learning curve, and limited budget. As Marc said, “You’re not going to need to buy a model from a model company… it just works with low hallucinations.”

Where are the mega ERP and other industry software players?

While chief commercial and marketing officers play pivotal roles in a company’s success, the bulk of any company’s business processes and expenses lies in areas dominated by SAP, Oracle, Workday, and a host of industry-specific solutions like commercial banking platforms, insurance underwriting products, and MRPs. Finance, procurement, and industry-specific operations are awash with opportunities for agentic AI solutions—or even services as software options. However, Oracle and SAP lack a vision. They are massive functionality laggards, and most industry-specific solutions are still tinkering with first-generation solutions such as DXC Technology’s Assure Answers for their insurance industry products. With nowhere else to go, business leaders are looking to a sea of niche and custom GenAI solutions—those that could threaten the dominance of major software players.

While the software companies in these areas are similarly slow-moving, the pace of change is fast. Also, there’s every reason to believe there will be a mass delivery of capability or third-party solutions that will extend creaky ERPs into the agentic AI era over the next 12–24 months. When this happens, business leaders will have a choice of using easy-to-implement, embedded GenAI capabilities or continuing to use their complicated and expensive DIY solutions.

Should I ‘skunkworks’ my GenAI solutions or leverage embedded GenAI capabilities?

When it comes to using GenAI, enterprises have a plethora of choices now. But enterprise personas can be categorized into three basic themes (see Exhibit 1), each with their own benefits and risks.

Exhibit 1: Three Models of Company GenAI Implementation

Source: HFS Research, 2024

Extenders will have two great concerns

Intellectual property (IP) terms should be the largest concern for Extenders. Typical software licensing IP terms grant all functionality ownership to software companies, leaving buyers solely with rights to their data. This made tremendous sense when legacy software configuration was based on workflows, business rules, and custom field integrations performed using built-in administrator capabilities. In the GenAI world, advanced computer learning capabilities are part of the core solutions, and it’s unclear how the ownership of learned capabilities created by customers based on their unique strategies and business rules will be firewalled from competitors.

Extenders should also remain greatly concerned with increased pricing. Already overburdened with massive software licensing fees, agent-driven solutions will come with transactional costs. While Salesforce has expressed that it will be flexible with its MSRP pricing of $2.00 per transaction when purchased in bulk, business owners will no doubt be troubled with paying the same amount for easy and complex transactions, defining the start and end of complicated transactions (which may result in multiple smaller transactions), and simply understanding the budget impacts of turning on Agentforce when volumes are not known. We anticipate some buyers are going to be surprised by their invoices.

The Bottom Line: For CIOs who have bet on Salesforce, betting on Agentforce as an embedded GenAI solution is a safe move to get more from it.

Surely, there will be examples of all three operating models within organizations—and maybe blends of them. However, the complexity and costs of creating and integrating your own GenAI capabilities are serious problems. It’s clear that business owners will prefer the easy path of implementing embedded technology by flicking a few switches and writing a few prompts rather than depending on IT departments to develop custom hyperscaler solutions or waiting for a small, relatively feature-weak LLM’s CEO to stumble on to their LinkedIn profile to propose a meeting.

While this path has its own governance and complexity risks, business owners are great at overcoming these speed bumps. The capabilities will soon be so fully embedded into software suites that it will be harder to skip them than just using them. Once Oracle and SAP wake up to the power of agentic AI and embed the technology in their core suites, finance, supply chain, and human resource leaders will rapidly turn on these configurations. The game will then be over for small-fry LLMs and the thousands of pilot investments companies have made.