Self-improving AI: what changes for B2B SMEs

The timeline: from Salesforce AI to 650 million for self-building AI

Richard Socher is not a new name in the artificial intelligence landscape. In the past, he led Salesforce's AI division and founded you.com, a search engine based on language models. In May 2026, however, it made a significant leap in scale. Its new startup raised $650 million with a stated goal: to build an AI system capable of self-search and self-improve indefinitely .

According to reports by TechCrunch , Socher emphasizes a crucial point. The project is not purely academic. Therefore, the company intends to release concrete commercial products, thus distinguishing itself from the foundational research of OpenAI or DeepMind in their early stages.

This detail is relevant. In fact, the gap between frontier research and enterprise applications is shrinking rapidly. Consequently, B2B SMEs can no longer afford to watch these developments from a distance.

Who wins and who is at risk: a no-nonsense read

Self-improving AI systems — also known as recursive self-improvement systems — represent a distinct category from current generative models. Instead of being trained once and then deployed, these systems modify their own parameters or architecture in response to new data and goals.

Among the main beneficiaries of this technology are, first and foremost, large enterprises with mature data infrastructures. They can leverage self-optimizing systems to continuously reduce operational costs. Furthermore, technology vendors that integrate these capabilities into their SaaS products will gain a structural competitive advantage.

Conversely, SMEs that depend on single suppliers risk finding themselves in a position of technological dependence. In fact, if the system managing their processes evolves autonomously, operational control shifts progressively towards the vendor. This is a concrete risk, not a hypothetical one.

Similarly, there are compliance risks. Self-modifying systems are hard to audit. Therefore, in regulated contexts — such as finance, healthcare, or logistics — they could create friction with regulations like the European AI Act .

SHM Studio's perspective: between enthusiasm and pragmatism

We at SHM Studio we have been following the evolution of AI systems applied to SMEs for several years. Our take on this news is nuanced.

On one hand, Socher's announcement confirms a direction that was already visible. Current generative models show capabilities of in-context learning increasingly sophisticated. Therefore, the shift toward systems capable of structural self-improvement is a logical progression, not a sudden quantum leap.

On the other hand, however, the gap between a 650 million announcement and a usable product for a manufacturing SME in Brescia or a B2B distributor in Turin remains considerable. Therefore, the operational advice is not to wait for this technology to start your digital transformation. On the contrary, it's the exact opposite.

Companies that build a solid data foundation, digitized processes, and in-house AI skills today will be the best positioned to adopt self-improving systems when they become available on the market. For this reason, investing in AI solutions applicable today is not in contradiction with future-proofing — it is its precondition.

Three concrete scenarios for Italian B2B SMEs

It is useful to translate this scenario into realistic use cases. Below are three contexts where self-improving systems could impact Italian B2B SMEs in the next 24-36 months.

• Adaptive commercial automation: CRM systems that autonomously optimize outreach sequences based on response rates, without human intervention. This continuously lowers the cost per qualified lead. The LinkedIn campaigns could benefit from self-calibrating AI layers.
• SEO and self-optimizing content: editorial platforms that automatically rewrite content based on ranking signals. Therefore, the role of strategic copywriting shifts towards editorial supervision and goal setting.
• Paid campaign management: systems that reallocate budgets across channels in real time, optimizing not just individual campaigns but the entire media architecture. This directly impacts the management of google ads campaigns and channel mix.

In all three cases, the value is not in the automation itself. In fact, the real value lies in the company's ability to define clear objectives, monitor results, and correct course. Therefore, human skills do not disappear—they shift to more strategic levels.

The construction site is still open: what is missing before mass adoption

Despite investor enthusiasm, significant technical and regulatory hurdles exist. First and foremost, self-improving systems require enormous computational infrastructure. This makes them, at least initially, accessible only to large organizations or via specialized vendor APIs.

Secondly, the issue of interpretability remains unresolved. As MIT Technology Review has documented, current models are already difficult to audit. A system that modifies itself adds another layer of opacity. Therefore, compliance with the European AI Act — which requires transparency and traceability — becomes a technical challenge even before it is a legal one.

Finally, there's the issue of organizational trust. Italian SMEs have historically adopted new technologies cautiously. Therefore, even when these systems are technically available, the adoption cycle will be longer compared to Anglo-Saxon markets. This isn't necessarily a disadvantage: it allows observing others' failures before investing.