As we accelerated into 2026—the fourth year in what can truly be described as the generative AI revolution—it became clear that this technology is transforming far more than just the tools we use; it is reshaping the very structure of how we work, create, and think. The arrival of agentic AI in 2025 marked a defining milestone, unleashing systems capable not only of producing content but of executing complex, self-directed tasks that previously demanded substantial human labor. Among these innovations, the explosion of agentic coding tools stood out. Applications such as Gemini Jules, Claude Code, and OpenAI Codex quickly redefined what was possible for programmers, enabling them to develop fully functional products from abstract prompts in only hours.
When I personally tested these systems—using OpenAI’s Codex to design four distinct WordPress plug‑in security modules, and Claude Code to craft a complete iPhone application—I witnessed firsthand how far the field had advanced. Work that once required countless days of effort and debugging was now achievable within a handful of focused sessions. These experiences set the foundation for a yearlong reevaluation of my professional AI toolkit and the cost-benefit relationship behind every subscription and upgrade.
To maintain complete transparency in my evaluations, I not only test these platforms but also purchase every professional plan myself, ensuring that no corporate influence shapes my conclusions. Thus, as 2025 began, my AI roster consisted of three essential tools: Midjourney for image generation, ChatGPT Plus for conversational intelligence and analysis, and Adobe Creative Cloud for comprehensive creative production. Together these amounted to about $100 per month, accumulating to $500 by the close of May—$350 of which was attributed solely to Adobe. Each filled a distinct function within my daily operations: Midjourney ($10 monthly) empowered visual storytelling projects; ChatGPT Plus ($20) provided reliable data analysis and access to superior language models; and Adobe Creative Cloud ($69.99) continued as my primary suite for high‑precision image and design work.
Midway through the year, I expanded my experimentation into areas of automation and data handling. By June, a self‑hosted archival system—Karakeep—was up and running. This project replaced my dependency on Mozilla’s Pocket and integrated OpenAI’s API for intelligent keyword extraction from nearly 21,000 stored articles. The server ran continuously for several months, generating approximately $35 in processing charges due to OpenAI’s token‑based billing. Though modest in cost, it was an invaluable improvement to my research workflow. With this setup, I still maintained my baseline expenses for Midjourney, ChatGPT Plus, and Adobe, leading to a cumulative expenditure of $835 by the conclusion of August.
September introduced a new phase: the age of what I have begun referring to as ‘vibe coding’—a process through which an AI agent autonomously constructs or refactors code under the user’s supervision. Utilizing OpenAI’s Codex tool as part of ChatGPT Plus, I quickly realized that even within the usage restrictions of the lower payment tier, I could compress what would normally be twenty full days of manual programming into just twelve hours of guided interaction. Unsurprisingly, these throttling restrictions soon convinced me to trial ChatGPT Pro at $200 per month. With unrestrained access, I succeeded in building four complete WordPress extensions in only four days—a feat equating to multiple years of ordinary progress. Recognizing this as an exceptional but temporary development sprint, I subsequently reverted to the more economical Plus tier once that surge of work concluded.
Around this period, I added Notion AI to my toolset, integrating it mainly for organizational automation, especially database generation based on structured lists. The $20 monthly business plan expanded my AI‑assisted productivity environment, raising my cumulative total to $1,234.90 by October, of which $700 still corresponded to Adobe.
November was defined by a shift of attention toward Anthropic’s Claude Code suite. While Codex demonstrated impressive capabilities, the developer community’s enthusiasm for Claude prompted me to evaluate it directly. I soon discovered why it had so rapidly gained favor: operating natively through terminal applications, Claude Code combined flexibility with accessible pricing. I began on its $20 Pro tier but swiftly upgraded to the Max 5x plan ($100 monthly) to accommodate heavier workloads. Over the span of seventeen days—using it for roughly eleven of those—I built a robust, feature‑packed iPhone application to manage my 3D‑printing filament inventory. The speed and reliability mirrored what I had observed earlier with Codex, reaffirming that AI‑driven development had reached a professional-grade maturity. As of November’s end, my running total stood at $1,455, with Adobe accounting for $770.
By December, it became impossible to ignore the disproportionate share of costs attributed to Adobe Creative Cloud—roughly half of all my AI spending for the year. Because I relied heavily on Photoshop but rarely opened Illustrator or the ancillary tools, I decided to scale back. Instead of the full $70 monthly bundle, I transitioned to the $20 Photography plan that still includes core Photoshop access. Adobe’s plan restrictiveness, including its limit to two devices and small allocation of generative credits, remains frustrating, but the savings were considerable. For comparison, 4,000 credits accompany the full plan, whereas the Photography plan only provides twenty‑five, necessitating occasional top‑ups via the Firefly Standard add‑on at $10 per 2,000 credits.
To continue assessing competitors for an upcoming article on AI image platforms, I briefly tested Canva Business—costing $20 for a single month—and its linked Leonardo AI engine. Despite their accessibility, the generative results failed to match my expectations, leading to their prompt cancellation. I also subscribed to Google’s AI Pro tier ($20) to test both Gemini 3 and the experimental Nano Banana Pro model. Though initial trials of Nano Banana’s object replacement functions were striking, ChatGPT’s newly released image capabilities began to reach similar proficiency, prompting me to treat the Google plan as a short‑term exploratory expense.
Ultimately, the December balance sheet summarized a consistent pattern: $30 in combined fees for Midjourney and ChatGPT Plus, $20 for Notion AI, $100 for Claude Code, $20 for Photoshop, and various transient costs for testing Canva and Google services. The month’s total rested around $210, culminating in an annual outlay of $1,665—with $790 traceable to Adobe-related spending.
Reflecting on these figures, I admit that my expenditure seems substantial at first glance. Nevertheless, it’s a deliberate investment tied directly to my role as a technology reviewer. Paying for each plan out of pocket ensures that every assessment remains unbiased and credible. Were I an everyday user rather than a professional evaluator, I would certainly trim back; yet, the reasoning behind maintaining certain key subscriptions stands firm. ChatGPT Plus remains indispensable for data analysis, mathematical verification, and research management. Claude Code and Codex have proven themselves transformative for accelerating software creation. Meanwhile, although Midjourney’s function could theoretically be replicated through other models, its artistic output remains uniquely imaginative and surprisingly affordable.
As for Notion AI, my usage is intermittent, but when called upon, it dramatically accelerates database construction and organizational tasks, more than justifying its fee. In retrospect, I clearly should have optimized my Adobe configuration far earlier. Still, Photoshop remains so tightly woven into both my creative rhythm and professional history that abandoning it entirely would harm productivity.
Taking all this into account, if I had minimized certain commitments sooner—namely the Adobe plan and the temporary coding tool upgrades—my 2025 spend would likely have hovered closer to the $900–$1,200 range. Yet, considering the sheer number of hours reclaimed, the expense proves defensible. Productivity tools that save measurable time essentially pay for themselves when compared against any reasonable hourly valuation of labor.
For readers evaluating their own subscriptions, the same principle applies: assess whether the time savings and cognitive relief genuinely offset the monetary costs. My own safeguard remains vigilant budgeting, assisted by my wife and business partner, who reviews our accounts daily. Deactivating unused tools immediately minimizes waste and prevents overlooked subscription creep. In an era when digital services proliferate without restraint, such discipline is critical.
Ultimately, the broader question extends beyond my own spend: how do you, as a creator or professional, determine which AI investments merit continuation? Have you seen tangible boosts in efficiency, or have some novelties faded with time? In this rapidly evolving ecosystem of agentic assistants, coding companions, and art generators, measuring real value has become an essential professional habit—one that separates mere curiosity from sustainable, intelligent innovation. For further discussions and updates on these experiments, you can continue to follow my posts and social channels, where I document daily progress and insights from the ongoing intersection of technology, creativity, and human ingenuity.
Sourse: https://www.zdnet.com/article/which-ai-tools-are-worth-paying-for-here-are-subscriptions-im-keeping-and-why/
