As we near the close of 2025, artificial intelligence has once again transformed the technological landscape far beyond what we anticipated just a year ago. When the curtain fell on 2024, we believed we understood the trajectory of generative AI and its implications for daily work and creative industries. Yet 2025 ushered in a dramatically more sophisticated era, driven primarily by a new phenomenon known as agentic AI—systems capable not merely of responding to human input, but of autonomously taking initiative, making decisions, and producing complex outputs across domains once considered firmly human.
Among the most powerful manifestations of this shift has been the explosive rise of agentic coding tools. Applications such as Gemini Jules, Claude Code, and OpenAI Codex now possess the ability to draft entire suites of software, ranging from small plug‑ins to fully developed applications, with minimal direct human instruction. To test these capabilities firsthand, I integrated both Codex and Claude Code into active development projects: four plug‑in security add‑ons for WordPress produced entirely by Codex, and an extensive iPhone application brought to life through Claude Code’s advanced functionality. These experiments vividly demonstrated how AI can compress what once required months or even years of traditional programming into mere hours or days.
Throughout the year, I broadened and revised my personal ensemble of AI tools, adding, refining, or eliminating subscriptions to determine which offerings yielded measurable improvements. In this detailed retrospective, I chronicle not only the platforms I retained, but also those that failed to justify their cost as we transition into 2026. Importantly, my analysis is completely independent: every commercial version of software discussed here was purchased out of pocket, ensuring that no vendor exerted influence over my evaluations.
At the dawn of 2025, my toolkit comprised three essential paid services: Midjourney, ChatGPT Plus, and the Adobe Creative Cloud suite. This initial lineup brought my monthly AI expenditure to roughly one hundred dollars, totaling five hundred dollars by the end of May—an investment heavily skewed toward Adobe products. Midjourney, which I first subscribed to in early 2023, quickly became my visual imagination partner. Its generative art capabilities allowed me to produce tailored imagery for a range of creative endeavors, from promotional designs for my wife’s online store to original cover art supporting my musical projects. ChatGPT Plus occupied a more analytical role. The upgrade from the standard tier unlocked advanced processing features, including data analysis and access to next‑generation reasoning models, turning the chatbot into an invaluable research and coding companion. Adobe’s suite, meanwhile, represented the legacy software backbone of my workflow. Photoshop and Illustrator, now augmented with enhanced generative capabilities, transformed familiar creative processes into AI‑accelerated production pipelines.
By midyear, my experimentation took a more technical turn. In June, I harnessed the OpenAI API to support a personal project: a self‑hosted content management server—my functional replacement for Mozilla’s now‑retired Pocket platform. This custom system, named Karakeep, automatically tags and categorizes saved articles through AI‑driven keyword extraction. The open‑ended flexibility of the API proved invaluable, though not inexpensive; between June and August, roughly thirty‑five dollars were consumed by token usage, bringing my cumulative 2025 AI spending to eight hundred thirty‑five dollars.
Autumn introduced a new chapter: the era of what I call “vibe coding.” This experimental practice involves supervising an autonomous code‑writing agent that executes software development tasks semi‑independently. Initially, OpenAI’s Codex was available via the ChatGPT Plus tier, but its throttled limits soon became evident. Even so, within a mere twelve hours of interaction, I achieved the equivalent of approximately twenty days of manual development. To overcome capacity restrictions, I temporarily upgraded to the $200‑per‑month ChatGPT Pro plan—a decision that immediately paid dividends. In only four days, the AI produced four separate add‑ons for my existing WordPress security suite, matching four years of previous productivity. After completing the project, I prudently reverted to the lower‑cost Plus tier to maintain fiscal control.
During this same period, I integrated Notion AI on its $20 Business plan, automating the generation of structured databases directly from raw data lists—an augmentation that streamlined my content‑organization process. By the end of October, these combined investments brought my total AI budget to roughly $1,235, with Adobe continuing to account for the lion’s share.
In November, professional curiosity led me to explore Anthropic’s Claude Code, which had swiftly become the darling of many development communities for its balance of cost efficiency and usability. Unlike Codex, Claude Code seamlessly functioned within a terminal interface, providing immediate hands‑on control. I subscribed to the service at the $20‑per‑month tier but soon upgraded to the $100 Max 5x plan after quickly depleting the basic token allowance. Across a focused seventeen‑day development sprint, I built a bespoke iPhone app designed to manage filament inventory for 3D printing in my fabrication lab. This experience not only validated Claude Code’s coding efficiency but also inspired plans for Mac and Apple Watch counterparts. By the close of November, my cumulative AI expenditure had climbed to $1,455, a direct reflection of my deepening experimentation.
When December arrived, I turned a critical eye toward my bloated Adobe expenditures. Historically, roughly half of my entire AI budget had gone toward maintaining the full Creative Cloud subscription—an excellent suite, but one whose breadth I rarely utilized. I made the pragmatic decision to downgrade from the $70 plan to Adobe’s more streamlined $20 Photography plan, preserving Photoshop access while shedding the rarely touched extras. In parallel, I briefly explored Canva’s Business plan and its integration with Leonardo AI for an additional $20, though their generative features fell short of my creative expectations, leading me to cancel after a single billing cycle. To round out the month’s trials, I subscribed to Google’s $20 AI Pro tier to evaluate Gemini 3 and Nano Banana Pro for forthcoming reviews.
When the financial dust settled, my December outlay stood at approximately $210, capping the year’s full AI expenditure at $1,665—of which $790 had ultimately flowed to Adobe. Evaluating these numbers underscores how easily subscription‑based innovation can inflate costs without sustained awareness.
Would I have spent this much had my job not involved continuous technology testing and editorial analysis? Almost certainly not. My personal philosophy remains one of frugality and intentional spending—an approach shaped by years of monitoring subscription creep across the digital economy. Nevertheless, I also assign considerable value to anything that saves significant time or cognitive effort. The hours reclaimed through AI‑assisted coding, image generation, and data management far outweigh their monetary costs when considered against the productivity returns.
I am ambivalent, however, about some tools’ long‑term necessity. Midjourney’s rich creative output still outpaces most competitors, and at ten dollars a month, it remains a bargain for its distinctive aesthetic power. Notion AI, too, earns its place in my workflow despite infrequent use—its sporadic interventions have rescued me from tedium during complex data structuring tasks. The real financial misstep was my delayed Adobe downgrade; shedding that expense months earlier would have reduced my total AI spending by nearly a third.
Ultimately, across 2025, my expenditures reflected not only curiosity but the tangible pursuit of efficiency. Had I optimized sooner—paring unnecessary subscriptions and limiting premium tiers to active development periods—my outlay might have hovered closer to $1,200. Excluding agentic coding tests entirely could have reduced that figure to roughly $900. Yet these investments were not wasted; they revealed how profoundly AI can transform creative and professional productivity. The thousands of cumulative hours saved convert directly into value, confirming that in both journalistic and operational terms, the technology has more than paid for itself.
For anyone managing multiple AI services, I recommend conducting similar audits: maintain transparency over where money flows, track renewal cycles, and cancel or downgrade immediately once usage patterns decline. My own household practice—daily account reviews with my wife and business partner—provides not only cost oversight but also immediate protection against unauthorized or redundant charges. In a world increasingly defined by automated subscriptions, this vigilance ensures balance between innovation and sustainability.
As time becomes our most limited resource, the equation often comes down to whether the minutes reclaimed through automation outweigh the dollars invested. In my case, working nearly seven days a week, the convenience and acceleration delivered by AI tools continue to justify their cost—at least for now. Looking ahead to 2026, the real challenge will lie not in discovering new AI capabilities, but in mastering discernment: knowing precisely which technologies genuinely elevate our work, and which simply dazzle without delivering enduring value.
Sourse: https://www.zdnet.com/article/ai-tools-subscriptions-productivity-coding-adobe-cost-break-down/
