Opinion article inspired by the report “The State of Enterprise AI” from OpenAI: how AI is becoming the invisible infrastructure of businesses, why a gap is widening between pioneers and laggards, and how Bubble chose as early as 2023 to operate as a “AI company” by fully relying on AI agents, automation, and internal copilots, both in the product and in the organization (and even in personal life).
Planned publication date: January 15, 2026
Overview
By 2025, enterprise AI is no longer a gadget. It is quietly becoming the invisible infrastructure of many organizations: copilots for teams, automation of repetitive tasks, assistants embedded into products, and agents that orchestrate entire workflows.
OpenAI’s report “The State of Enterprise AI” highlights three clear facts:
- adoption is exploding and use cases are getting deeper (Custom GPTs, Projects, APIs, agents);
- companies that take it seriously see real gains in time, productivity, and revenue;
- a gap is opening between leaders and laggards.
At Bubble, this doesn’t come as a surprise. Since 2023 we have made a fairly radical choice for a small French fintech: to behave as if we were already a future company, and to organize our work around new AI tools instead of trying to bolt them onto a 2010-style organization.
This article covers:
- how AI is reshaping the modern enterprise;
- why a large execution gap is opening between leaders and everyone else;
- how, very concretely, at Bubble we try to stay on the frontier – in the product, in the way we work, and even in our personal lives.
1. The new enterprise AI race
1.1. From experimentation to core workflows
For years, “enterprise AI” mostly meant:
- a POC in a corner;
- a flashy demo for the executive team;
- a marketing chatbot on the website.
OpenAI’s report shows that phase is ending. We now see:
- an explosion in messages sent via ChatGPT by enterprise users;
- an increase of 320× in reasoning token consumption per organization on the API;
- growing use of Custom GPTs, Projects, and agents as actual business building blocks.
In other words, AI is leaving the lab and moving into real operational processes.
1.2. Tangible gains, not vague promises
The numbers are telling:
- most enterprise users report saving 40–60 minutes per active day with AI;
- in some roles (data, engineering, communications, finance), that climbs to 60–80 minutes;
- use cases range from faster IT incident resolution and campaign execution to data analysis, coding, and regulatory or research monitoring.
What matters is not just raw productivity. It is the fact that technical tasks are changing hands: non-technical people are starting to write small scripts, automate spreadsheets, prototype tools, and build their own internal GPTs.
1.3. A global, cross-industry movement
Another important signal: adoption is no longer confined to U.S. tech companies.
- Countries like Australia, Brazil, France, and the Netherlands are growing faster than the global average.
- Sectors like healthcare and manufacturing are rapidly catching up with tech and finance.
So the question is no longer “will AI transform my industry?”, but rather:
How ready is my organization to benefit from it, and how fast?
2. Leaders vs laggards: a widening gap
2.1. Same tools, radically different usage
One key point from the report:
- frontier workers (roughly the top 5% of users) send several times more messages than the median user;
- frontier firms consume far more intelligence (credits, reasoning tokens, advanced tools) than everyone else.
On paper, everyone has access to similar models. In practice, we see:
- some people opening ChatGPT once a week to “try it”;
- and others who have made it a permanent work reflex: writing, research, coding, analysis, synthesis, documentation, project support, and more.
The report is clear:
Gains scale with depth of use.
The more task types you use AI for (writing, analysis, coding, images, research, translation…), the more hours per week you save.
2.2. Business impact is already visible
The studies cited in the report show that AI‑mature companies:
- grow revenue faster;
- improve margins;
- innovate more (patents, new products);
- and often see higher employee satisfaction.
We are well past the “nice to have” phase. When integrated deeply into systems, data, workflows, and culture, AI becomes a structural advantage.
3. What this means for Bubble: building a AI company, since 2023
When we started Bubble in 2023, we made a fairly simple decision:
If we want to build a next‑generation product, we also need to build a next‑generation company.
Not a company where AI is just used to generate prettier LinkedIn posts, but one where:
- AI agents, automations, and internal copilots are central to how we operate;
- we are willing to challenge “normal” ways of working inherited from traditional firms;
- what we say in our marketing is first a reflection of our daily practice, not the other way around.
3.1. In the product: a genuinely agentic portfolio co-pilot
On the product side, this shows up in Bubble Portfolio:
- a conversational agent that chains screening → backtests → allocations → order-file generation;
- a modular architecture (screener, backtest engine, multi-factor scoring, risk optimizer, broker connectors) designed to be extensible;
- an agentic logic: the system does not just answer a question, it orchestrates a process, while keeping the user in control.
Our compass here is simple:
- no black box;
- no custody of client accounts;
- no AUM-based fee model;
- transparent subscription pricing, around €0–10 per month (example) for access to the co-pilot, independent of portfolio size.
3.2. In the organization: agents, automation, and documentation everywhere
The way we work internally follows the same principles.
A few concrete examples:
- Research and monitoring: we rely heavily on agents to aggregate reports, articles, scientific papers, regulatory updates, and then we filter, challenge, and synthesize.
- Analysis and prototyping: we combine backtests, notebooks, simulations, and AI to explore strategies, test hypotheses, and model scenarios.
- Operational automation: whenever a workflow is repetitive, we try to turn it into an agent-orchestrated process (data quality, scoring, reporting, execution follow-up, etc.).
- Documentation: we document as much as possible in Notion and then use it as external memory for our own internal agents.
The goal is not to “replace” people, but to allow a tiny team to operate as if it were 10× larger, while staying lucid about model limits and keeping the final say.
3.3. In our personal lives: coherence before storytelling
This choice does not stop at Bubble as a product. We are already living a large part of this “2025 company” mindset in our personal lives:
- we use agents to organize projects, learn faster, manage our information diet, and prepare important decisions;
- we test new tools very early, often on our own problems first, before bringing them into the product;
- we try to stay honest about what really works, what breaks, and what genuinely saves time – or doesn’t.
This is not a marketing storyline. It is simply how we operate, with all the strengths, limits, and contradictions that come with it.
4. What other companies can learn from the report – and from our approach
Not every company needs to become an AI-first fintech. But many can draw similar lessons.
4.1. The challenge is no longer to “try AI”, but to structure it
Useful questions to ask:
- Is AI confined to POCs, or integrated into critical processes?
- Are key datasets accessible via APIs and actually used in real workflows?
- Do business teams have time, space, and permission to experiment with internal GPTs and agents?
4.2. Everyone has models. Not everyone has discipline
What distinguishes leaders in the report is:
- real governance (security, compliance, quality);
- a willingness to document, factorize, and reuse what works;
- a progressive but determined approach: start small, measure, harden, industrialize.
4.3. Learn from others, without copy‑pasting
Our context is specific (fintech, portfolios, multi‑broker, regulation). But the underlying idea – “build a 2025 company, not just a 2025 product” – is portable:
- to an industrial SME automating quality control or maintenance;
- to a consulting firm turning its knowledge into internal copilots;
- to a hospital or insurer building better guidance through a complex system.
The key is not to do “what everyone else does,” but to decide:
Which parts of our organization deserve to be redesigned in light of AI, not just sped up?
Conclusion: the real question is no longer “if”, but “how”
“The State of Enterprise AI” confirms what we see every day:
- AI is becoming a core layer of the modern enterprise;
- a widening gap is emerging between companies that integrate it deeply and those that stay at the surface;
- the upside is not just productivity, but new products, new experiences, and new business models.
At Bubble, our bet is straightforward:
- build a portfolio co-pilot that is truly aligned with investor interests;
- and build, in parallel, an organization that already behaves like a AI company – staying on the frontier of innovation without losing our compass: decency, transparency, alignment.
If this resonates with you, the best way to form an opinion is the same we use internally:
- start small;
- put an agent into a real workflow;
- measure what changes;
- then decide whether you, too, want to start building … – today.
