Artificial intelligence has grown shockingly rapidly from a niche experiment to a mainstream business priority. As we enter 2025, virtually every industry is grappling with how to use AI for competitive advantage. Business leaders and even the general public are inundated with information – but which sources provide the most reliable, insightful statistics on where AI stands today?
In this article, we review five standout AI reports published in 2024–2025, each offering a unique perspective on the state of AI adoption and its impact.
If you're a leader trying to assess what other businesses are doing, a worker trying to anticipate how your role will change, or part of a technical team exploring how AI is impacting IT & operations, one or all of these studies will help you.
The five reports we’ll explore are:
McKinsey – “Superagency in the Workplace: Empowering People to Unlock AI’s Full Potential at Work”
Glide – “State of AI in Operations 2025”
Writer – “Enterprise Generative AI Adoption Survey”
Vellum – “State of AI 2025”
Retool – “State of AI 2024”
Each report is data-rich and detailed, and each is aimed at uncovering a specific type of AI insight. We highly recommend you click through or download the full reports if they seem applicable to your business.
These reports cover everything from workforce readiness and operational deployment to enterprise strategy, development hurdles, and the evolving AI tech stack. We’ll break down key statistics and insights from each study, explain what they mean for businesses, and highlight what each report is best for. By the end, you’ll have a clear, big-picture view of where AI trends are headed.
McKinsey – Superagency in the Workplace: Empowering People to Unlock AI’s Full Potential at Work
McKinsey’s “Superagency in the Workplace” is best for understanding leadership challenges in enterprise AI adoption and diving deep into detailed data.
It provides hard data on the mismatch between C-suite perceptions and employee reality, highlighting that cultural and organizational factors are the main bottlenecks. Business decision-makers will find value in its discussion of how to lead an AI-focused digital transformation, align the organization, and empower people with AI. If you’re an executive wondering why AI isn’t delivering expected value yet, this report offers a candid look in the mirror – and guidance to turn things around so that your company can better use AI’s full potential at work.
Key stats and insights from McKinsey
McKinsey’s January 2025 report focuses on how enterprises can scale AI in the workplace. A striking finding is that almost all companies are investing in AI, yet only 1% feel they have achieved “AI maturity” – meaning AI is fully integrated into their operations.
In fact, over the next three years 92% of companies plan to increase AI investments, but virtually none consider themselves fully scaled up in AI usage. Why the gap? McKinsey concludes that the biggest barrier isn’t employee resistance – employees are largely ready – but a lack of leadership drive and vision. Leaders are not steering fast enough in adopting AI at scale, even though workers are eager to embrace these tools.
One revealing statistic compares perceptions: in McKinsey’s survey, 94% of employees and 99% of C-suite executives report at least some familiarity with generative AI tools, showing widespread awareness. However, executives dramatically underestimate how much their people already use AI. Leaders guessed only 4% of employees are using AI for a substantial portion of their work, when in reality about 13% of employees self-reported doing so – over three times more than bosses thought.
Similarly, nearly half of employees (47%) believe that AI could handle at least 30% of their jobs within a year, whereas only 20% of executives think employees will reach that level of AI usage so soon. This optimism gap suggests that frontline staff see the potential of AI to transform work, but management may be playing catch-up in recognizing and planning for it.
Implications
The McKinsey report suggests that employees are not the ones dragging their feet – on the contrary, many workers are already experimenting with AI and expect it to lighten their workload significantly. The implication is that leadership needs to step up by setting a clear AI vision, investing in training, and accelerating deployment. The finding that only 1% of companies feel “mature” in AI usage despite significant investment indicates many firms are stuck in their adoption process. Short-term returns on AI can be unclear, but the long-term $4+ trillion productivity potential is enormous.
Leaders need to bridge this gap by moving from experimentation to full integration of AI into workflows. Otherwise, they risk falling behind competitors. The report emphasizes that the risk for leaders “is not thinking too big, but rather too small” when it comes to AI ambition. In other words, hesitancy and incrementalism at the top are holding organizations back more than employee readiness or technology limitations.
Glide – State of AI in Operations, 2025
Glide’s State of AI in Operations 2025 report is best for understanding the operational rollout of AI, benchmarking your company’s progress against others, and finding guidance for your own AI adoption process.
It provides concrete stats on adoption rates, investment levels, and industry-by-industry progress. If you’re a manager or executive wondering how your business compares with competitors, or looking for reassurance that AI can deliver real operational benefits, this report offers valuable perspective. It’s particularly useful for identifying common roadblocks in implementation (knowledge, culture, legacy tech) and for learning how early adopters (especially in tech) are using AI effectively.
Learn about the state of AI in 2025
Read the full reportKey stats and insights from Glide
Glide’s AI in Operations 2025 report zeroes in on the practical operational adoption of AI across companies. Published in early 2025, it surveyed over 1,000 managers and senior business leaders to gauge where businesses stand in implementing AI in their day-to-day operations.
The findings show that AI adoption is well underway and accelerating. According to the survey, 28% of businesses are already actively using AI, and another 45% have active plans to implement AI, meaning almost 73% of companies are engaged with AI to some degree. (Only a small minority, 8%, said they have no interest in adopting AI.)
This indicates that AI is being adopted at a faster rate than perhaps any technology in recent memory. As Glide’s report points out, it took Google over a decade to reach a billion daily searches, but ChatGPT hit a billion daily queries in just two years – a testament to AI’s rapid uptake.
Not only are many companies rolling out AI, but they’re also overwhelmingly optimistic about its impact. In the Glide survey, 87% of business leaders expected AI to have a positive impact on operations, with 28% predicting transformational impact and 59% anticipating a somewhat positive impact. What’s more, early adopters are finding the impact even greater than expected: of the businesses that have already implemented AI, 51% report that AI has been transformational to their operations in practice, exceeding the initial hype. In other words, for about half of those using AI, the technology is truly improving workflows (e.g. automating tasks, enabling better decision-making), not just providing minor improvements.
Companies are also putting their money behind their optimism. Glide found that businesses using AI plan to increase investment in 2025; 57% of companies already using AI plan to spend over $100,000 on AI initiatives in the next year (and one-third of companies in the planning stage also expect to invest at least six figures). This budget trend shows that once firms see results from initial AI projects, they tend to double down and allocate even more resources to AI.
We also see a divide by industry and role: tech companies and IT teams are at the forefront. In fact, 72% of tech sector companies have already deployed AI “agents” (automated AI systems) in their operations. These might be things like AI-driven customer support chatbots, automated workflows, or other AI tools integrated into software. Tech firms also report higher success with AI, likely due to greater familiarity and easier integration.
On the other hand, more traditional industries – construction, manufacturing, retail – are moving slower. Their top barriers include a lack of knowledge into AI, internal resistance or conflict, and difficulty integrating AI with legacy systems. In short, every industry is interested in AI, but some face practical hurdles in catching up with the tech sector’s pace of adoption.
Implications
The Glide report’s data suggests that AI is not just hype—it’s already delivering real value on the ground. For business leaders, the fact that a majority of peers are either using or actively implementing AI should be a signal that we’ve passed the “if” stage and are now in the “how” stage of AI adoption. Companies that delay may find themselves lagging behind competitors who are using AI to cut costs, speed up operations, or improve customer experiences.
The strong optimism (nearly 9 in 10 expecting positive outcomes) and the finding that actual impact can exceed expectations (51% seeing transformational change) imply that the payoff of well-executed AI projects can be substantial. However, the differences across industries also carry an important implication: organizations in sectors with implementation challenges may need to invest in training (to address knowledge gaps) and modernizing infrastructure (to handle AI tools) as foundational steps.
The barriers like internal conflict hint that change management is crucial – getting teams aligned and comfortable with AI. Also, seeing tech and IT teams lead adoption means that many companies start AI implementation in technical or digital departments before expanding elsewhere. Business leaders in less digitally mature sectors might consider partnering with third-party experts, agencies, or outside consultants to overcome initial hurdles. Overall, the Glide report paints a positive picture: rapid uptake and generally positive results. But it also reminds us that those positive results depend on overcoming skills gaps and integration issues, especially outside the tech bubble.
Writer – Enterprise Generative AI Adoption Survey
Writer’s Enterprise Generative AI Adoption Survey focuses specifically on generative AI (versus agentic AI) use by knowledge workers at the enterprise level. It is best for understanding the human and organizational side of generative AI adoption in large companies.
This report is a valuable resource for executives who want to foresee and navigate the cultural challenges of rolling out generative AI at scale – things like turf wars between departments, employee skepticism, and the need for change management. The report is rich with statistics that can help CIOs, CTOs, and CEOs benchmark their progress (e.g., are we investing enough? are we seeing typical ROI or falling behind?).
It’s also useful for identifying success levers: the survey makes a strong case that having a formal strategy, committed leadership, and internal champions can make or break an enterprise AI program. In short, Writer’s report provides a reality check on enterprise AI implementation: it’s not all rosy, but it offers guidance on how to turn challenges into opportunities so AI can truly transform the business rather than cause chaos.
Key stats and insights from Writer
The Writer “Enterprise Generative AI Adoption Survey” (2025) provides a candid look at how large organizations are integrating generative AI (think GPT-style tools) and the internal dynamics around it. Conducted with Workplace Intelligence, it surveyed 1,600 U.S. executives and knowledge workers, offering both top-down (C-suite) and bottom-up (employee) perspectives.
96% of organizations in the survey say AI is a “key enabler” for their company’s future, indicating near-universal belief in the importance of generative AI. Additionally, almost 97% of companies anticipate expanding AI usage to new departments (such as HR, training, and customer support) in the near future. In other words, virtually every enterprise is not only continuing with AI but spreading it across the organization. This indicates that business leaders are on board with AI.
However, the report reveals that the road to adoption has been bumpy. A significant 2 out of 3 executives (around 66%) say introducing generative AI has led to tension or power struggles within their company, and 42% even fear that AI adoption is “tearing the company apart” due to internal divisions. These are startling numbers that highlight internal friction: AI initiatives have triggered debates over who “owns” AI (IT vs business units), how resources are allocated, and which strategies to pursue. In fact, the survey notes specific rifts – about two-thirds of C-suite respondents report tension between IT teams and other business units over AI implementation, and 71% say various groups are building AI solutions in silos rather than a coordinated effort. This fragmentation can lead to duplicated efforts and inconsistent strategies.
Within teams, shadow IT is growing, with more than one-third (35%) of employees paying out-of-pocket for AI tools they want to use at work (because their company isn’t providing suitable tools), and 31% of employees – especially younger staff – admit they actively resist or sabotage AI initiatives (for example, refusing to use AI tools or trust AI-generated outputs). Clearly, there’s a disconnect: while leadership is sold on AI, on the ground, people may feel fearful (of job impact or AI errors) or frustrated (by lack of proper tools), leading some to covertly use AI or push back against it.
Another mixed outcome is in ROI. Companies are investing heavily – 73% of companies said they’re spending at least $1 million per year on generative AI – yet only about one-third have seen significant ROI so far. Despite big budgets, many have not recouped value at scale, which might be behind some of the “massive disappointment” voiced by 1 in 3 executives about AI projects. It’s not that AI isn’t delivering any value (in fact, the vast majority of both workers and execs using AI said they have personally benefited from generative AI tools, and at least 90% are optimistic about their company’s AI approach). Rather, the expectations were too high, and early implementations may not yet be living up to the hype, especially if poorly coordinated.
The Writer survey also explored what can be improved. Companies that invest more and approach AI systematically do better. For example, having a formal AI strategy makes a huge difference. At organizations with an organization-wide AI strategy, 80% of executives reported their AI adoption as very successful, versus only 37% at companies without a clear strategy. That’s a 2x gap in success rates, emphasizing the importance of top-down direction and planning.
Additionally, the report highlights the role of “AI champions” at organizations. Among employees using AI, 77% are identified as AI champions or potential champions – meaning they are enthusiastic adopters who could help drive AI efforts if empowered. Nearly all of these champions (98%) are willing to help build or refine AI tools for their company, and 94% have seen career benefits from engaging with AI. This suggests companies should tap into these eager users to pilot projects, train colleagues, and evangelize AI internally.
Another point: higher investment correlates with better outcomes. This might seem obvious, but the data quantified it – there’s roughly a 40 percentage-point difference in success between the highest and lowest AI investors. This implies that half-measures may not yield the desired ROI; a significant commitment of resources (budget, talent) is often needed to get the full potential impact of AI. Lastly, executives overwhelmingly (98%) feel that choosing the right AI vendor or platform is critical, and many are not fully satisfied with their current vendors, indicating they expect more support and partnership from AI solution providers.
Implications
The Writer survey indicates having enthusiasm for AI isn’t enough – you need the right strategy and solid change management. Enterprises clearly believe in AI’s promise (nearly everyone sees it as key), but without alignment, generative AI projects can cause infighting and disappointment.
The findings suggest a few actionable implications:
Develop a clear AI strategy and communicate it. This means getting IT and business on the same page, defining use cases, governance, and goals for AI so that efforts aren’t siloed. It also means addressing employee concerns – for example, providing the AI tools employees find useful (so they don’t feel the need to BYO tools) and reassuring staff through training and transparency (to reduce resistance driven by fear).
Empower AI champions in your workforce. Those who are excited about AI can be enlisted to pilot new solutions, train peers, and demonstrate wins, helping to build credibility and encourage adoption.
Be prepared for iterative progress on ROI. The fact that many haven’t seen big ROI despite spending suggests that AI adoption is a learning process – initial projects might not get immediate results, especially if experimentation is needed. Companies should measure outcomes and iterate, rather than expecting instant transformation. Over time, as the organization learns and strategies improve, ROI can grow.
Vellum – State of AI 2025 (AI Development & Deployment)
Vellum’s State of AI 2025 report is best for understanding the technical side of AI development – where companies are in building AI solutions and what hurdles they encounter.
It’s particularly useful for technology leaders (CTOs, product managers, heads of R&D) or innovation-focused executives who want to benchmark their progress in developing AI against industry trends. If you are wondering “Are we behind or ahead in our AI buildout? What problems should we anticipate?”, Vellum’s data offers guidance.
It focuses on the technical and operational challenges that might not appear in more business-oriented surveys – things like model hallucination rates, use of dev tools, and evaluation methods. It’s also forward-looking, giving a sense of what the next-generation AI projects will focus on (customer-facing apps, AI agents, etc.). Vellum provides a developer’s-eye view of AI in 2024–25, which complements the higher-level business perspectives of the other reports.
Key stats and insights
The Vellum “State of AI 2025” report approaches AI from the perspective of AI builders and developers. It surveyed over 1,200 tech professionals (often in startups or R&D roles) to understand how far along companies are in developing and deploying AI solutions, and what challenges they face. The results show that as of end of 2024, most organizations are still in relatively early stages of AI development.
Only 25.1% of surveyed companies have actually deployed an AI application in production (either for customers or internal use). The rest are in various stages of the journey: about another 25% are still in the strategizing phase (planning or figuring out what to build), ~21% are building a proof-of-concept, roughly 14% are beta-testing with users, and the remainder are in very early exploration (gathering requirements or evaluating a proof-of-concept).
This finding balances some of the hype – despite the explosion of AI interest in 2023–2024, many companies have not fully deployed AI yet. It aligns with the notion that AI adoption is widespread, but full implementation takes time. Even large enterprises are often still in pilot phases for many AI initiatives.
When it comes to development hurdles, the Vellum report provides a clear ranking of what trips teams up the most. The number one challenge (cited by 57% of respondents) is handling AI “hallucinations” and prompt management. Hallucinations refer to AI models confidently generating incorrect or nonsensical information – a common issue with generative AI. Engineering AI prompts (how questions/commands are given to the AI) and outputs to reduce these errors is a primary headache for AI builders.
Right behind that, 42.5% said prioritizing the right use cases is a major challenge. With AI capable of so many things, it can be tricky for companies to identify which project will have the most impact or is most possible, leading to analysis paralysis or diffused efforts. The third biggest challenge, for 38%, is lack of technical expertise – essentially a talent gap. Skilled AI engineers and data scientists are in high demand, and not every organization feels it has the right people to build workable AI solutions.
Other notable challenges include model performance (speed and scalability), noted by 33%, and data access/security concerns (ensuring AI has the data it needs, without compromising privacy or compliance), noted by 32%. Interestingly, “securing buy-in from key stakeholders” was lower on the list, at about 21%, suggesting that at least among these respondents, getting leadership or client support for AI projects is less of an issue than the technical and strategic hurdles. The dominance of technical issues like hallucinations and performance indicates that the technology’s limitations are very much front-of-mind for builders – they know AI can be powerful, but making it reliable, accurate, and fast enough for production use is hard work.
Despite these challenges, most AI teams are adopting practices to improve their outcomes. One encouraging stat: over 57% of developers already perform systematic evaluations of their AI models, and another 31% plan to start doing so. That means nearly 9 in 10 are either testing their AI’s accuracy/quality regularly or intending to – a good sign of maturing development processes. Only 12% said they do no AI performance evaluations at all.
Looking forward, Vellum asked companies about their plans for AI in 2025. The top priority, cited by 59%, is to build more customer-facing AI use cases. This suggests that while internal or back-end AI applications have been the safer starting point (as noted in other reports too), companies are now keen to develop AI features or products that directly interact with customers (think AI-driven customer service, personalized recommendations, etc.). The second priority, from 55% of respondents, is to develop more complex “agentic” workflows. “Agentic AI” refers to AI systems that can autonomously perform multi-step tasks or make decisions (sometimes described as AI agents).
Companies want to create AI that doesn’t just respond in a single step (like answering a question), but can carry out a process (for example, an AI agent that could take a support ticket from intake to resolution by querying databases and executing actions). The bigger trend is businesses moving from simple chatbot or single-task AI toward more sophisticated, automation-heavy AI systems.
Recognizing that to use AI effectively, they need people who understand AI better 42% of businesses plan to invest in upskilling their teams in AI skills. Many also intend to build their organization’s own AI solutions for internal use (38%) and/or use third-party AI tools to improve internal operations (33%). A smaller subset (17%) said hiring more AI developers is a 2025 plan, indicating hiring isn’t the top strategy compared to upskilling existing staff or leveraging external tools. Overall, these plans show an ambition to scale AI efforts – more projects, more complexity, more integration into products and daily work.
Implications
For business and tech leaders, the Vellum report is a reminder that developing AI solutions is a journey with many stages, and most companies are still in early to mid stages.
If your organization hasn’t deployed a production AI application yet, you’re in good company – three-quarters of others haven’t either. Companies shouldn’t be complacent. They should understand that getting to AI products that show ROI takes time and iteration. On the other hand, being among the first 25% to go live with AI could be a competitive advantage in your industry if you manage it wisely.
The breakdown of challenges provides a checklist for project planning: Expect to invest in solving data quality and AI accuracy issues (like hallucinations), and ensure you have or can hire the right technical talent. It’s interesting that prioritizing use cases is a top challenge – this implies leaders should spend more effort in the strategy phase, clearly linking AI projects to business value. A well-chosen pilot use case (one that’s impactful but feasible) can build momentum, whereas a poorly chosen one can cause frustration.
Lastly, this report highlights a trend: 2025 will likely see AI being more public and more autonomous. Companies are planning customer-facing AI and agentic workflows, indicating a push towards AI that can handle more complex tasks. Leaders should start thinking about this today. How will you ensure an AI agent makes good decisions? How will customers react to more AI in services?
Retool – State of AI 2024 (Real Use Cases and the AI Stack)
Retool’s State of AI 2024 report is best for benchmarking the astronomically fast progress of AI in the last year against the other four newer reports, specifically for technical teams.
AI is developing faster than any other technology in history. This Retool report from 2024 provides a detailed overview of AI adoption in technical teams that stands in sharp contrast with the findings of the 2025 reports.
The report is very useful for technology and innovation managers who want to know “What use cases are working? What tools are others using? Where are we on the adoption curve?” This report cuts through hype by focusing on real use (e.g., how many are using ChatGPT daily, what problems they encounter). For a business leader, it provides insight into which functions in your company might benefit most from AI (the data says we should look at product development, engineering, customer support, and operations), and how your tech teams might be feeling about AI (probably that they could do even more with it). It’s also great for understanding the developing AI software ecosystem – who is important (OpenAI, etc.), and how companies are creating AI solutions (use of APIs, databases, model fine-tuning).
Key stats and insights from ReTool
Retool’s State of AI 2024 report offers a practical snapshot of how technical teams (developers, data scientists, product managers, etc.) were using AI in the first half of 2024. This report focuses on AI use cases, user sentiment, and the tools enabling AI in organizations.The survey included about 750 tech professionals across various company sizes and roles.
One aspect Retool examined is AI adoption maturity and investment. They found that only a smallnumber of companies considered themselves at the leading edge: in 2024, just about 10% of respondents described their company as highly advanced in AI adoption, down from roughly 13% who said that in 2023. Overall roughly 30% total felt they were moving quickly – while the rest felt they were at a slower pace. This indicates a bit of a reality check compared to the previous year, possibly as companies recognized how complex deploying AI can actually be.
Importantly, almost nobody thinks they’re overspending on AI. Only 4.5% of those surveyed felt their company is investing “too much” in AI, whereas the vast majority feel their investment is either “just right” (42%) or “not enough” (40.5% say their company should invest more in AI). This suggests pent-up demand: employees see opportunities for AI that aren’t being fully funded or explored yet. It’s implies to management that tech teams are generally hungry for more AI resources, not less.
On individual usage, Retool discovered that AI tools have become routine for many workers. Over half (56.4%) of respondents use generative AI tools like ChatGPT almost every day at work, and nearly all (90%+) use them at least weekly. Interestingly, this daily usage was highest in the smallest companies – at startups with 1–9 employees, about 72% were using AI daily, versus around 50% in larger firms (and dipping to ~43% in mid-size 1000–5000 employee companies). The likely reason: smaller companies often have more flexibility and fewer strict policies, so employees freely experiment with new tech.
By role, product managers and engineers topped the charts with 68% and 63% using AI almost daily, whereas designers were lowest around 39%. This shows that AI has been embraced most in technical and product development functions (to speed up coding, analysis, content drafting, etc.), and less so in design or perhaps administrative roles.
Those who use AI frequently report significantly higher productivity gains. Among daily AI users, 64.4% said AI has substantially improved their productivity; compare that to only 17% of weekly users and just 6.6% of occasional users who reported big productivity boosts. This correlation implies that the more comfortable and integrated AI is in one’s workflow, the more value it yields – perhaps because people learn how to use it better, and AI can handle more tasks when used consistently. For business leaders, it hints that encouraging regular use and providing help is a good idea.
Another interesting insight is “shadow AI.” In 2023, Retool found that about one-third (34%) of employees were using AI tools at work in secret–without their management’s knowledge or approval. In 2024, this number improved but was still noteworthy: 27.3% of respondents admitted they use AI at work secretly. Over a quarter are “hiding” their AI use. Why hide it if AI is so beneficial?
The survey investigated the reasons and found that they often weren’t because of explicit bans—only 9% of secret users were actually violating a company policy by using AI. More commonly, the secrecy was a result of unclear company guidelines, fear of how it would be perceived, or internal politics. Some employees worry that if they use AI openly, it might be frowned upon (maybe their boss doesn’t trust AI outputs, or they fear it might imply their own work is replaceable).
The good news is the trend is moving towards openness–the secret usage percentage declined from the prior year. Many organizations are developing AI usage policies now, with about 63% of respondents saying they are paying attention to AI regulations and company policies. This suggests that by clarifying acceptable use and addressing concerns, companies can make AI more open and effective.
When it comes to specific use cases, Retool’s data shows some shifts between late 2023 and mid-2024. Most common use cases stayed roughly similar in popularity, but two trends were interesting: customer support chatbots and workflow automation saw roughly 5 percentage point increases in adoption while coding assistance and copywriting saw small declines. The share of teams using AI for writing code dropped from 47.5% to 42.1%, and for content copywriting from 32.9% to 28%. Meanwhile, usage of AI for support chatbots rose from 28.9% to 33.9%, and using AI to automate internal workflows rose from 12.9% to 17.8%.
AI is changing. Early on, much AI use was individual-focused (help me write code or marketing copy faster). As we moved into 2024, more teams started deploying AI to interact with users (chatbots) or streamline processes (automation), which are more organizational implementations. However, it was noted that despite the popularity of AI chatbots as a use case, not all customer support teams are using AI heavily yet—likely due to caution and trust issues.
For pain points and trust, Retool found that the top challenges in developing AI applications mirrored those found by Vellum: accuracy and data. Specifically, the two most cited pain points were AI model output accuracy/hallucinations (38.9%) and data access/security (33.5%), which stayed consistent with their previous survey. When asked how much they trust AI outputs, respondents on average gave it a lukewarm 6.1 out of 10 confidence. That implies moderate trust at best – people know AI can be wrong, so they aren’t blindly relying on it.
This cautious stance explains why many companies prefer internal AI uses first. In fact, only 8.5% of respondents saw greater promise in external (customer-facing) AI than internal. 33.7% saw more promise in internal applications, and the remaining ~58% saw equal potential in both. In practice, this means a lot of firms are first using AI internally (e.g., decision support tools for employees, internal knowledge assistants, etc.) where mistakes are contained, rather than exposing customers to AI outputs too early.
Finally, the Retool report dives into the AI tech stack – which AI technologies and platforms are teams actually using. OpenAI’s models utterly dominate current usage. Around 76.7% of AI users in the survey rely on OpenAI models in production, with OpenAI’s GPT-4 being the single most-used model (45% of respondents use it) and GPT-3.5 second (25%). This isn’t too surprising given the popularity and API availability of those models.
However, alternatives are growing – Anthropic’s Claude model made gains (its usage share roughly quadrupled compared to earlier, with the release of Claude 2/Claude 3), and new models like Mistral showed up in responses (Mistral is a newer open-source model). Users were relatively happy with their AI models/providers overall: about 35.5% were very or mostly satisfied, 34.5% somewhat satisfied, and only around 17% were actively unhappy enough to want to switch providers.
When it comes to tailoring AI, a majority of organizations are doing some form of model customization rather than using it out of the box. About 29.3% of teams are fine-tuning existing models on their own data, and 23.2% are using vector databases or retrieval augmentation to give AI custom knowledge. Larger enterprises (5,000+ employees) use retrieval techniques even more (around 33%), likely because they have a ton of proprietary data to integrate. Companies are investing in feeding AI with the right data or tweaking it to improve relevance and accuracy.
Certain tools have significantly increased: vector database usage jumped from only 20% of respondents in 2023 to 63.6% in 2024. Vector databases (like Pinecone, Weaviate, etc.) are specialized databases for storing embeddings that enable semantic search – a core part of retrieval-augmented generation (RAG) pipelines. This massive jump indicates that organizations are discovering that connecting LLMs to their own knowledge base helps to prevent hallucinations and give more factual answers.
About 17.7% of companies are building their own AI models from scratch. While still a minority, it’s significant—nearly 1 in 5. This suggests that some companies (especially AI product companies or those with very specific needs) are investing in developing proprietary models despite the high cost and expertise required.
Implications
The Retool report, while technical and already beginning to be outdated, still gives practical insights for business leaders.
First, AI usage is active in daily workflows, particularly in tech-facing roles – and those who use it more, get more value. Encouraging broader, regular use of AI (with support and oversight) can boost productivity across teams. If your product managers and engineers aren’t using AI yet, they might be missing out on the efficiency gains their competitors are getting.
The shift in popular use cases – towards customer support and operations automation – indicates that AI is moving from a personal productivity hack to an enterprise tool for customer service and business processes. For leadership, this means AI projects are scaling up in scope. It might be time to invest in AI platforms that integrate with customer support systems, or RPA (robotic process automation) combined with AI for workflow improvements.
The big rise in vector databases and retrieval tech underscores a simple lesson: combining AI with proprietary data is essential. Generic AI knowledge isn’t enough; companies want AI to know their products, their policies, their content. Business leaders should ensure their AI teams have a strategy for knowledge management (what data to feed the AI, how to keep it updated, etc.).
The big takeaways on AI for business leaders
Across these five reports, a consistent picture emerges of a world where AI is no longer experimental – it’s becoming an integral part of business, albeit with some growing pains. A few overarching trends stand out:
Widespread adoption with varied maturity: Virtually all companies are now investing in AI at some level, and most have active projects or plans underway. Yet, only a tiny fraction (around 1%–10%, depending on the survey) consider themselves fully mature or “leading” in AI. The majority are still feeling their way through pilots, strategy-building, or early deployments. This means opportunity abounds for those who can move from proof-of-concept to full production – there’s a chance to leapfrog competitors if you can execute well.
Optimism about impact, but realism about challenges: Business leaders and workers overwhelmingly believe AI will have positive or even transformational effects on operations and growth. This optimism is fueling increased investment and rapid experimentation. However, experiences in 2024 balanced the hype with some hard truths. Key challenges like model accuracy (hallucinations), data integration, and talent gaps came up repeatedly. Organizational challenges – from leadership inertia to internal turf wars – have also emerged as critical issues to address. The lesson: AI’s promise is real, but it’s not a magic wand. It takes careful implementation, the right people, and often new processes to unlock value.
Employees are eager (and already using AI): One refreshing insight is that employees at all levels are jumping on AI tools, sometimes faster than their companies. Surveys show high individual usage of tools like ChatGPT for daily work, and many workers feel AI could offload a sizable chunk of their tasks soon. In fact, some employees are self-servicing their AI needs – whether quietly using AI behind the scenes or even paying for their own AI software when the company doesn’t provide it. This shadow IT is a double-edged sword: it means your workforce is ready, but if you don’t provide support and guidelines, there may be inconsistency or security risks. Smart organizations will take advantage of this energy by guiding employees through training on AI best practices, official tool rollouts, and encouraging knowledge-sharing.
Leadership and strategy make the difference: A recurring theme is that leadership commitment and clear strategy are key to turning AI experiments into enterprise-wide success. Companies with strong executive vision and formal AI strategies are seeing much better outcomes. Leadership needs to proactively address internal friction–aligning with IT, involving stakeholders across departments, and setting out governance for AI use. Collectively, these reports suggest that AI adoption is as much about organizational change as it is about technology.
From generative AI to agentic AI: In 2023–24, many organizations focused AI efforts on internal use cases or productivity tools. Now, as we head into 2025, there’s a noticeable shift toward customer-facing AI and more sophisticated AI applications. Whether it’s deploying AI chatbots for customer support, personalized AI-driven services, or “agentic” AI systems that can perform complex tasks autonomously, companies plan to push AI further into core products and direct customer interactions.
The developing AI tech stack: Several technical trends are making this AI jump possible. Large language models (LLMs) from providers like OpenAI remain widely used, but the ecosystem is expanding with new models and open-source alternatives. Companies are increasingly customizing their AI tools with the help of consultants, AI developers on staff, and no code platforms that enable them to customize their AI interfaces.
The big takeaway: 2025 is poised to be a year where AI moves into a period of deeper implementation.
The companies that benefit most will be those that can integrate AI into their operations and offerings effectively, which requires investing in people (skills and culture) as much as in technology. The data shows that the potential rewards (productivity gains, transformative impact, competitive edge) are there, and, importantly, that early adopters are largely seeing positive outcomes. But realizing those rewards universally will require navigating the challenges highlighted in these reports – from finding, customizing, and deploying the right tools, to managing change within the organization, to building trust in AI within your team.
Sources:
