# THE MIDDLEMAN TRAP ## When Your AI Strategy Runs on Someone Else's Engine ### A Case Study in Creative Destruction — Adobe, Firefly, and the $200 Billion Moat That Wasn't **Jeep Marshall** LTC, US Army (Retired) Airborne Infantry | Special Operations | Process Improvement February 2026 --- > **Series Note:** This is Paper 4 in the *Herding Cats in the AI Age* series. Paper 1 ("The Super Intelligent Five-Year-Old") established that AI needs doctrine, not more intelligence. Paper 2 ("The Digital Battle Staff") showed the military already built the coordination frameworks the civilian AI industry lacks. Paper 3 ("The PARA Experiment") demonstrated those principles in a live Obsidian vault laboratory. This paper examines what happens when a $141 billion company tries to herd AI cats without doctrine — and discovers the cats belong to someone else. --- ## EXECUTIVE SUMMARY Adobe Inc. (NASDAQ: ADBE) spent $948 per customer per year selling creative professionals the most comprehensive software suite in the industry. Then generative AI arrived, and Adobe made a strategic decision that Wall Street is still punishing: instead of building the best AI engine, they became the best *showroom* for everyone else's engines. Adobe's Firefly platform now integrates models from Google (Gemini/Nano Banana), OpenAI (GPT Image), Black Forest Labs (FLUX), Runway, Pika, Ideogram, Luma AI, Moonvalley, ElevenLabs, and Topaz Labs. Adobe markets this as "multi-model choice." The market reads it differently: the company that defined digital creativity for 40 years now resells other companies' AI to its own customers through a markup wrapper. The stock tells the story. ADBE closed at $257.78 on February 28, 2026 — down 43% from its 52-week high of $464.33 reached just one year earlier.[^1] The all-time high of $688.37, set on November 19, 2021, now looks like a different era. Goldman Sachs issued a rare Sell rating. Jefferies cut its target from $400 to $290. The 52-week low of $251.10, hit on February 12, 2026, sits dangerously close to current trading levels.[^2] This paper applies the same analytical frameworks from Papers 1-3 — MDMP mission analysis, Lean Six Sigma process assessment, QASA quality review, and ASS2 security evaluation — to Adobe's AI strategy. The findings are not encouraging. Adobe is executing what this series identifies as the "Middleman Trap": a company that maintains the user interface but surrenders the value-creating engine to competitors who will eventually cut out the middleman. This paper was produced using a team approach: Lean Six Sigma Black Belt (process analysis), QASA Quality Standards (quality assessment), ASS2 Security (strategic vulnerability), and Creative Arts practitioners (field testing with real-world evidence). --- ## 1. THE THESIS: FROM CREATOR TO CURATOR The creative software industry is experiencing a phase transition. For four decades, Adobe's moat was technical superiority — Photoshop, Illustrator, Premiere Pro, and After Effects were not just market leaders but category definitions. You didn't edit photos; you "Photoshopped" them. That verb-as-brand status represents the deepest moat in software. Generative AI dissolved that moat in 18 months. The thesis of this paper is direct: **Adobe's pivot from tool creator to AI model curator represents a strategic surrender disguised as platform flexibility.** When your flagship AI product runs on Google's engine, your competitive advantage reduces to UI familiarity and enterprise contracts — both of which erode faster than Wall Street currently models. This thesis connects directly to Paper 1's core argument. Adobe's problem is not that AI models are insufficient. The problem is that Adobe lacks the operational doctrine to deploy AI as a force multiplier rather than a replacement for its own capabilities. They are herding cats — brilliant, fast, tireless cats — but the cats belong to Google, OpenAI, and Black Forest Labs. And half of them are chasing mice that don't exist. --- ## 2. ADOBE STOCK: THE MARKET'S VERDICT ### The Numbers | Metric | Value | Source | |--------|-------|--------| | Current Price (Feb 28, 2026) | $257.78 | NASDAQ real-time | | 52-Week High | $464.33 (Feb 18, 2025) | FinanceCharts | | 52-Week Low | $251.10 (Feb 12, 2026) | FinanceCharts | | All-Time High | $688.37 (Nov 19, 2021) | MacroTrends | | YTD Decline | -25.58% | FinanceCharts | | 12-Month Decline | -43.88% | FinanceCharts | | FY2025 Revenue | $23.77 billion (+11% YoY) | Adobe 10-K | | Forward P/E | 15.06x | StockAnalysis | | Analyst Consensus | Buy (21 analysts) | StockAnalysis | | Average Price Target | $409.14 (+58% upside) | StockAnalysis | | Goldman Sachs Rating | Sell ($290 target) | 24/7 Wall St. | ### The Disconnect Here is the paradox that defines Adobe's current position. Revenue grew 11% to a record $23.77 billion. Net margins hit 30%. AI-influenced annual recurring revenue exceeded $5 billion — up from $3.5 billion in 2024. More than 35% of Photoshop subscribers actively use generative AI features. By every traditional software metric, Adobe is executing.[^3] Yet the stock lost 43% in twelve months. Goldman Sachs issued one of its rare Sell ratings. HSBC cut its target from $388 to $302. Jefferies dropped from $400 to $290.[^4] The market is pricing something the earnings reports don't yet show: the structural risk that Adobe's AI integration strategy accelerates its own obsolescence. Every partner model Adobe brings into Firefly teaches customers that the value lives in the AI engine, not the Adobe wrapper around it. ### Historical Price Trajectory The 40-year stock history reveals three eras. Era 1 (1986-2010): Desktop publishing dominance. Era 2 (2011-2021): Cloud subscription transformation driving the stock from $30 to $688. Era 3 (2022-present): The AI reckoning, with the stock giving back 63% of its all-time high as the market reprices Adobe for a world where its moat may not hold.[^5] --- ## 3. THE MIDDLEMAN EVIDENCE: A FIELD TEST ### What We Did On February 28, 2026, we conducted a controlled experiment using Adobe Firefly's web interface. We generated identical images using the same prompt through two different models available within Firefly: Adobe's native Firefly Image model and Google's Gemini 2.5 Flash Image (Nano Banana), accessed through the partner model dropdown menu inside Adobe's own platform. **The Prompt:** *"A pack of sleek black cats with glowing cyan bioluminescent circuit-board markings"* This prompt was chosen deliberately — it is the visual identity of this paper series ("Herding Cats in the AI Age") and tests both artistic capability and text rendering, the two domains where AI image generation most directly competes with Adobe's traditional tools. ### What We Found **Adobe Firefly Native Model Output:** - The image produced beautiful, atmospheric cyberpunk cats with glowing circuit-board markings - Artistic quality was high — moody lighting, depth of field, strong composition - The text rendered as garbled nonsense: "BEARETIXSLUGE" and "PA'TEXCACT LEFIMENT" - No readable text was produced despite the prompt context requiring it **Google Gemini (via Adobe's Own Platform) Output:** - The image rendered clean, legible text: "HERDING CATS IN AI AGE" - Strong structural composition with four distinct cats - Circuit-board markings were more geometric and consistent - Overall image was more "designed" and less "artistic" — but it actually worked for its intended purpose ### The Wireframe Test We repeated the experiment with a technical prompt: *"A wireframe architecture diagram of a PARA method knowledge vault (Projects, Areas, Resources, Archives)"* **Firefly Native:** Produced an artistic isometric diagram with a pink color scheme. The PARA labels were partially readable but the sub-labels degraded into noise. Visually interesting but functionally useless as an actual architecture diagram. **Gemini Through Firefly:** Produced a structured top-down wireframe with clearly labeled PARA categories, hierarchical sub-folders, and file names. Some lower-level text degraded, but the overall architecture was functionally communicative. A designer could use this as a starting point. ### What This Means Adobe's own interface demonstrates that Google's model outperforms Adobe's native model on functional creative tasks. A customer paying $948/year for Creative Cloud can, inside Adobe's own product, compare Adobe's AI against Google's AI — and watch Google win. This is the Middleman Trap in action. Adobe built a showroom where customers comparison-shop the competition inside Adobe's store. --- ## 4. SWOT ASSESSMENT ### Expert Panel This SWOT analysis was developed by a cross-functional team applying frameworks from this paper series: - **LSS Black Belt (Process):** Lean Six Sigma analysis of Adobe's operational model - **QASA (Quality):** Quality standards assessment of Firefly output - **ASS2 (Security):** Strategic vulnerability assessment of partnership dependencies - **Creative Arts Practitioner:** Field assessment of real-world creative workflow impact --- ### STRENGTHS **S1. Market Entrenchment (LSS Assessment)** Adobe's Creative Cloud remains the industry standard. The file format ecosystem (.PSD, .AI, .INDD, .AEP) creates switching costs that no competitor has overcome. Enterprise procurement contracts, training infrastructure, and workflow integrations represent a process architecture that took decades to build. From a Lean perspective, these are deeply embedded value streams that competitors must either replicate or circumvent — not easy when your entire creative team's muscle memory runs on Adobe's keyboard shortcuts. **S2. Revenue Engine Intact (LSS Assessment)** FY2025 revenue of $23.77 billion with 11% growth and 30% net margins demonstrates that the subscription model still generates cash. AI-influenced ARR at $5 billion represents real monetization. The machine is still printing money even as the market narrative turns bearish. Remaining performance obligations of $22.5 billion provide forward visibility that most software companies lack. **S3. Commercially Safe AI (QASA Assessment)** Adobe's Firefly models are trained exclusively on licensed content — Adobe Stock, public domain, and expired-copyright material. This provides genuine IP safety for enterprise customers. In a legal environment where AI-generated content faces uncertain copyright status, Adobe's "commercially safe" positioning is a real differentiator. Content Credentials (digital provenance tracking) add a trust layer that no AI startup currently matches. **S4. Multi-Model Platform Strategy (Creative Arts Assessment)** Adobe's 60%-of-creators-use-multiple-models research validates the platform approach. Offering Gemini, GPT Image, FLUX, Runway, and others inside one interface eliminates subscription juggling and context-switching for creators. The value proposition is real — if Adobe can maintain it without commoditizing their own role. --- ### WEAKNESSES **W1. Native Model Inferiority (QASA Assessment)** Our field testing confirmed what creators are discovering in production: Adobe's native Firefly models consistently underperform partner models on functional tasks. Text rendering, structural accuracy, and prompt fidelity all favor Google's Gemini and Black Forest Labs' FLUX over Adobe's native output. When 60% of creators report using multiple models, and the best models inside your platform aren't yours, you have a quality gap that no marketing can overcome. The garbled text ("BEARETIXSLUGE") in our Firefly native test is not a cherry-picked edge case — it represents a systemic weakness in Adobe's generative architecture. **W2. Value Migration to Partners (LSS Assessment)** Every time a creator selects Gemini or FLUX from the model dropdown, they learn that the AI engine — not the Adobe interface — is the value-creating component. This is a classic Lean "waste of inventory" problem: Adobe invested billions in Firefly development ($1.8B+ R&D in AI) but users increasingly treat the native model as the inferior option they skip past to reach the partner model they actually want. The process flow moves value downstream to partners rather than capturing it internally. **W3. Seat Compression Threat (ASS2 Assessment)** AI automates tasks that previously required multiple human operators. If one designer with AI tools can produce the output of three designers without AI, enterprise customers need fewer Adobe seats. This "efficiency paradox" directly attacks Adobe's per-seat subscription model. The strategic vulnerability is compounded because Adobe itself is accelerating the efficiency gains through its AI integrations — essentially helping customers reduce their Adobe spend.[^6] **W4. AI Monetization Gap (LSS Assessment)** Strong AI adoption metrics (35% of Photoshop users, 5B AI-influenced ARR) mask a process efficiency problem: the conversion rate from "AI feature usage" to "incremental revenue" remains unclear. Users engage with AI features but may not be upgrading plans or paying more per seat because of them. The gap between engagement and monetization is the Lean equivalent of high throughput with low yield. --- ### OPPORTUNITIES **O1. Enterprise AI Governance (ASS2 Assessment)** Adobe's Content Credentials and commercially safe model training create a defensible position in enterprise AI governance. As AI-generated content faces regulatory scrutiny (EU AI Act, US executive orders), Adobe can position itself as the compliance platform — not just the creative platform. Companies generating marketing materials, legal documents, and public communications need provenance tracking that no AI startup provides at Adobe's scale. **O2. Agentic Creative Workflows (LSS Assessment)** Adobe's October 2025 announcement of "agentic AI assistants" signals a pivot toward automated creative workflows — not just generation but end-to-end creative production with AI agents handling layout, formatting, localization, and distribution. If Adobe can build the orchestration layer on top of partner models (applying the MDMP principles from Paper 2), the platform becomes an AI C2 node rather than just a model showroom. This is the difference between being a middleman and being a commander. **O3. Creative Cloud Pro Consolidation** Adobe's new Creative Cloud Pro plan bundles access to both native and partner models under a single subscription. If pricing captures sufficient value from the partner model access, Adobe can potentially monetize the curation layer itself — charging a platform fee for model access rather than competing on model quality. **O4. WPP/Enterprise Partnerships** The February 2026 expansion with WPP (the world's largest advertising company) to integrate Firefly Foundry for custom, brand-safe generative models represents the high-margin B2B play. If Adobe becomes the platform through which Fortune 500 companies deploy custom AI models with enterprise governance, the consumer-grade model quality concerns become less relevant. --- ### THREATS **T1. Google's Direct Competition (ASS2 Assessment — Critical)** Google launched advanced image editing features in its Gemini app in February 2026, directly competing with Photoshop. This is the strategic nightmare scenario: Google's Gemini model already outperforms Adobe's Firefly inside Adobe's own interface. Now Google is removing the middleman entirely. Why pay Adobe $948/year to access Google's AI through a Photoshop wrapper when Google offers the same AI natively? The Benzinga report noted this announcement immediately pressured both Adobe and Figma shares.[^7] **T2. Anthropic's Automation Tools (ASS2 Assessment — Critical)** On February 3, 2026, Anthropic introduced automation tools that Trefis identified as a direct threat to Adobe's core business model. Claude's computer use capabilities, combined with its code generation and document creation features, enable non-designers to produce professional creative output without Adobe's tools. This paper itself was produced using Claude — including research, analysis, formatting, and Obsidian-optimized markdown — with zero Adobe tools in the pipeline.[^8] **T3. AI-Native Competitors (Creative Arts Assessment)** Midjourney, Canva, Figma, and Runway are not just feature competitors — they are architectural competitors. Born in the AI era, they built from the ground up around generative workflows rather than bolting AI onto 30-year-old desktop software architectures. Canva's $40 billion valuation and Midjourney's viral adoption among non-designers represent market creation, not market share theft. They are expanding the addressable market while simultaneously commoditizing Adobe's position within it. **T4. The Seat Count Collapse** Multiple analysts (Trefis, Goldman Sachs, Seeking Alpha) have identified the same structural threat: AI makes individual creative professionals dramatically more productive, which means enterprises need fewer of them, which means fewer Adobe seats. Even if revenue per seat increases through AI upselling, total seat count compression can overwhelm per-seat gains. One analyst called it the "SaaS Apocalypse" — the moment when AI efficiency gains turn from tailwind to headwind for subscription software.[^9] **T5. Open-Source Model Convergence** Black Forest Labs' FLUX is open-source. Stable Diffusion remains free. The quality gap between proprietary and open-source image generation models continues to narrow. If the AI engine becomes commoditized (as LLMs are trending toward with Meta's Llama and Google's Gemma), Adobe's entire "multi-model platform" strategy collapses because users can access the same models without paying Adobe's platform tax. --- ## 5. LSS BLACK BELT ASSESSMENT: THE PROCESS PROBLEM ### Value Stream Analysis Applying Lean Six Sigma value stream mapping to Adobe's creative workflow reveals a fundamental process restructuring underway. **Traditional Adobe Value Stream (Pre-AI):** Raw Concept → Photoshop/Illustrator → Designer Skill → Polished Output In this model, Adobe's tools were the *process* — the mechanism through which human skill became creative output. Every step required Adobe software. Every output embedded Adobe's value. **Current Adobe Value Stream (AI Era):** Text Prompt → AI Model Selection → Generation → Adobe Tools for Refinement → Output Adobe shifted from being the entire process to being one step in a process where the critical value-creating step (AI generation) is performed by a partner's engine. This is the Lean equivalent of a factory that outsourced its core manufacturing to a supplier and retained only finishing and packaging. ### Waste Identification (DOWNTIME Framework) | Waste Type | Adobe Manifestation | |-----------|-------------------| | **Defects** | Firefly native model produces garbled text, requiring re-generation or switching to partner models | | **Overproduction** | Multiple model outputs generated to compare quality, wasting compute and credits | | **Waiting** | Creators wait for generation across multiple models to find acceptable output | | **Non-Utilized Talent** | Professional designers forced into "prompt engineering" rather than design craft | | **Transportation** | Context switching between native and partner models within the same interface | | **Inventory** | Unused AI credits accumulating under plans where generation quality doesn't meet standards | | **Motion** | Repeated prompt iteration to overcome model limitations | | **Extra-Processing** | Generated outputs require extensive manual refinement in Photoshop — defeating the time savings AI promised | ### Sigma Level Assessment Based on our field testing and industry reports, Adobe Firefly's native model operates at approximately a 2-3 Sigma quality level for functional creative tasks (text rendering, technical diagrams, prompt fidelity). This translates to 66,800-308,500 defects per million opportunities — acceptable for artistic exploration but unacceptable for production creative work. Partner models through the same interface operate at a notably higher sigma level, particularly for structured outputs.[^10] --- ## 6. QASA QUALITY STANDARDS REVIEW ### Assessment Criteria The Quality Assurance Standards Assessment evaluates Adobe's AI strategy against four quality dimensions: **Output Quality:** Does the product consistently meet professional creative standards? - **Finding:** Mixed. Adobe's native Firefly models produce high-quality artistic imagery but fail on functional tasks requiring text accuracy, structural precision, or prompt fidelity. Partner models compensate but expose the native model's limitations through direct comparison. **Process Quality:** Does the production workflow minimize waste and maximize value? - **Finding:** Degraded. The multi-model selection process introduces decision fatigue and comparison overhead that did not exist in the pre-AI workflow. Creators report (per Adobe's own study) using multiple models per task — a process quality indicator that no single model meets the quality bar consistently. **Standards Compliance:** Does the product meet regulatory and industry standards? - **Finding:** Strong. Content Credentials, commercially safe training data, and IP provenance tracking represent industry-leading quality standards. This is Adobe's clearest competitive advantage. **Continuous Improvement:** Does the organization demonstrate systematic quality improvement? - **Finding:** Concerning. Adobe's strategy of adding more partner models suggests they are treating quality gaps as a selection problem rather than a capability problem. Adding Gemini 3.1, GPT Image 1.5, and FLUX Kontext to the dropdown doesn't improve Firefly's native quality — it papers over it with more options. ### Quality Verdict Adobe passes QASA standards on compliance and safety but receives a conditional finding on output quality and process efficiency. The recommendation is to invest in native model capability rather than expanding the partner model catalog — or explicitly reposition as a creative platform orchestrator rather than a creative AI developer. --- ## 7. ASS2 STRATEGIC VULNERABILITY ASSESSMENT ### Threat Matrix | Vector | Severity | Likelihood | Impact | Priority | |--------|----------|-----------|--------|----------| | Google removes Gemini from Firefly | Critical | Medium | Revenue model collapse | 1 | | AI-native competitor achieves feature parity with Creative Cloud | Critical | Medium-High | Seat count erosion | 2 | | Open-source models reach production quality | High | High | Platform value erosion | 3 | | Enterprise customers build direct AI integrations | High | Medium | Bypass Adobe entirely | 4 | | Regulatory action limits AI-generated content | Medium | Low-Medium | Could benefit Adobe's safe models | 5 | ### Strategic Dependencies Adobe's partnership model creates mutual dependencies with asymmetric risk profiles. Google does not need Adobe to distribute Gemini — Google has its own Gemini app, Android integration, Google Workspace, and YouTube. Adobe does need Google (or equivalent partners) to remain competitive in AI generation quality. This asymmetry is the core vulnerability. When you depend on a partner more than they depend on you, you have a negotiating disadvantage that compounds over time. Every Gemini improvement that delights Adobe's customers simultaneously strengthens Google's standalone competitive position. ### Kill Chain Analysis The worst-case scenario unfolds in three steps: **Step 1 — Capability Transfer:** Adobe educates its customer base that AI model quality, not Adobe software, determines creative output quality. (This is already happening through the model comparison workflow.) **Step 2 — Direct Access:** AI model providers launch consumer-friendly creative tools that bypass Adobe's interface entirely. (Google Gemini's image editing launch in February 2026 represents this step.) **Step 3 — Seat Erosion:** Enterprise customers reduce Adobe seat counts as AI-augmented designers produce more output per person, while individual creators migrate to AI-native tools that cost a fraction of Creative Cloud's $948/year price point. Adobe's own multi-model strategy accelerates all three steps. --- ## 8. CREATIVE ARTS PRACTITIONER ASSESSMENT ### Field Report As a practitioner using both Adobe Creative Cloud ($948/year) and AI tools across multiple platforms, the field reality is this: the creative workflow has fundamentally changed, and Adobe is on the wrong side of the change. **Pre-AI Creative Workflow (2020):** I open Photoshop. I use skills developed over years to create, edit, composite, and polish images. The skill is in my hands. The tool is Adobe's. Neither works without the other. **Post-AI Creative Workflow (2026):** I type a text prompt. An AI model generates an image. I refine it in Photoshop if needed — but increasingly, the generated output is sufficient for its intended purpose without Photoshop refinement. The skill is in the prompt. The tool is Google's (or OpenAI's, or FLUX's). Adobe's role reduces to optional post-processing. ### The Image Evidence The images generated for this paper series tell the story: **Figure 1: Adobe Firefly Native Model — "Herding Cats" Prompt** ![[Firefly_A pack of sleek black cats with glowing cyan bioluminescent circuit-board markings et 94009.png]] *Adobe's native Firefly Image model. Beautiful artistic rendering. Text output: garbled nonsense ("BEARETIXSLUGE"). Unusable for any context requiring readable text.* **Figure 2: Google Gemini via Adobe Firefly — Same Prompt** ![[Firefly_GeminiFlash_A pack of sleek black cats with glowing cyan bioluminescent circuit-board markings et 94009.png]] *Google's Gemini model, accessed through Adobe's own Firefly interface. Clean, legible text: "HERDING CATS IN AI AGE." Functional for design use.* **Figure 3: Adobe Firefly Native — PARA Architecture Wireframe** ![[Firefly_A wireframe architecture diagram of a PARA method knowledge vault (Projects, Areas, R 509149.png]] *Firefly's native model. Artistic isometric view. Labels partially readable but sub-text degrades. Interesting as art; useless as architecture documentation.* **Figure 4: Google Gemini via Firefly — Same PARA Wireframe Prompt** ![[Firefly_GeminiFlash_A wireframe architecture diagram of a PARA method knowledge vault (Projects, Areas, R 509149.png]] *Google Gemini through Firefly. Structured top-down wireframe with clear PARA hierarchy. Functional as a starting point for actual architecture documentation.* ### The Verdict from the Field When the best output from your platform comes from your competitor's engine, you are no longer a tool company. You are a distribution channel. Distribution channels survive only as long as the supplier needs them — and Google, OpenAI, and Black Forest Labs are building their own distribution faster than Adobe can build its own AI engines. --- ## 9. CONNECTING TO THE SERIES: WHY DOCTRINE MATTERS HERE Paper 1 argued that AI needs military doctrine — structured planning frameworks, quality gates, and operational discipline — more than it needs raw capability. Adobe's situation is the corporate case study for that argument. Adobe has capability. Record revenue. Strong margins. Billions in AI-influenced ARR. A platform with every major AI model integrated. The capability is not the problem. Adobe lacks doctrine. There is no visible strategic framework for how the multi-model approach creates sustainable competitive advantage rather than accelerating commoditization. There is no "commander's intent" that unifies the Firefly native development roadmap with the partner model integration strategy. There is no "course of action comparison" that war-games the scenario where Google pulls Gemini from Firefly or where OpenAI launches a creative suite. Paper 2 showed that the military solved multi-agent coordination through hierarchical C2 (command and control), standardized communication protocols, and doctrine that defines relationships between elements. Adobe's AI strategy operates without any equivalent doctrine. The partner models are agents with no commander's intent, no battle rhythm, and no unified operational picture. Paper 3 demonstrated that even a single practitioner's Obsidian vault — with proper PARA organization, Claude Code CLI integration, and structured protocols — can coordinate multiple AI agents effectively. Adobe, a $141 billion company with 29,000 employees, appears less doctrinally disciplined in its AI coordination than a retired Army officer's personal knowledge management system. That observation should alarm Adobe's shareholders more than any earnings miss. --- ## 10. THE OFF-RAMP: WHAT ADOBE CAN STILL DO ### Option A: Commit to the Platform Play Adobe explicitly repositions as the creative operating system — the orchestration layer that coordinates multiple AI engines rather than competing with them on model quality. This requires: - Abandoning Firefly native as a competitive AI model - Investing in orchestration, governance, and workflow automation - Pricing the platform on value-of-coordination, not cost-of-generation - Building the creative equivalent of the military C2 architecture described in Paper 2 **Risk:** Shareholders may not accept the margin compression from surrendering the AI engine race. The "platform fee" model is harder to defend than the "premium AI" narrative Adobe currently tells. ### Option B: Win the Engine Race Adobe doubles down on Firefly native development to achieve quality parity with Gemini, GPT Image, and FLUX. This requires: - Massive R&D investment in model training and architecture - Acquiring or licensing training data at scale beyond Adobe Stock - Competing directly with Google DeepMind, OpenAI, and Meta FAIR on model quality - Accepting that they are 18-24 months behind the leaders **Risk:** The leaders are not standing still. Google's Gemini 3 and OpenAI's GPT Image represent moving targets that Adobe's Firefly has not demonstrated the ability to match, let alone surpass. ### Option C: Own the Enterprise Governance Layer Adobe pivots from creative tool company to creative compliance company — the platform through which enterprises deploy AI-generated content with full IP provenance, regulatory compliance, and brand safety. Content Credentials become the product; Photoshop becomes the delivery vehicle. **Risk:** Regulatory moats are fragile. If the EU AI Act mandates provenance tracking for all AI-generated content, every platform must implement it — Adobe's early-mover advantage evaporates. ### Recommended COA Option A with elements of Option C. Adobe's sustainable advantage lies in orchestration and governance — not in winning the AI model race. The military learned this lesson: you don't need the best individual soldiers. You need the best coordination of capable forces. Adobe has the platform. They need the doctrine. --- ## CONCLUSION: THE CATS BELONG TO SOMEONE ELSE Adobe's situation crystallizes the central argument of this entire paper series. Intelligence without doctrine produces impressive demonstrations and unreliable operations. The brilliant cats — Google's Gemini, OpenAI's GPT Image, FLUX's photorealistic generation — are tireless, fast, and producing stunning work. But they don't belong to Adobe. They are running through Adobe's barn temporarily, and nobody at Adobe has told them where to stay. The stock price reflects this reality. The earnings reports don't — yet. When they do, the gap between $257 and $251 (the 52-week low) will not be the floor. It will be the starting point of a revaluation. Adobe has time. Record revenue and strong margins buy runway. But runway without doctrine is just expensive waiting. And in the AI age, waiting is the one strategy that guarantees failure. > "Scaling AI agents is herding cats. The cats are brilliant, tireless, and fast — but nobody told them where the barn is, and half of them are chasing mice that don't exist." Adobe built a beautiful barn. The cats just don't live there. --- ## DISCLOSURE This paper does not constitute investment advice. The author does not hold positions in ADBE. Analysis is based on publicly available financial data, published analyst reports, and hands-on product testing as a Creative Cloud subscriber. All AI image comparisons were conducted using Adobe Firefly's web interface with results documented through screenshots. --- ## APPENDIX A: IMAGE COMPARISON METHODOLOGY All images were generated on February 28, 2026, using Adobe Firefly's web interface (firefly.adobe.com). The model selection was made via the Model dropdown menu within Firefly's Text to Image module. Native Firefly Image model and Google Gemini 2.5 Flash Image (Nano Banana) were selected sequentially with identical prompts. No post-processing was applied to any generated image. Screenshots were captured at native resolution. **Prompt 1:** "A pack of sleek black cats with glowing cyan bioluminescent circuit-board markings" **Prompt 2:** "A wireframe architecture diagram of a PARA method knowledge vault (Projects, Areas, Resources, Archives)" Both prompts were entered verbatim with no additional style modifiers, negative prompts, or parameter adjustments. ## APPENDIX B: STOCK DATA SOURCES | Source | URL | Data Retrieved | |--------|-----|---------------| | Yahoo Finance | finance.yahoo.com/quote/ADBE | Real-time price, historical | | StockAnalysis | stockanalysis.com/stocks/adbe | Analyst consensus, targets | | MacroTrends | macrotrends.net/stocks/charts/ADBE | 40-year price history | | FinanceCharts | financecharts.com/stocks/ADBE | 52-week range, returns | | Benzinga | benzinga.com/money/adobe-stock-prediction | Analyst targets, forecasts | ## APPENDIX C: SERIES INDEX | Paper | Title | Focus | Status | |-------|-------|-------|--------| | 1 | The Super Intelligent Five-Year-Old | AI needs doctrine, not intelligence | Draft Complete | | 2 | The Digital Battle Staff | Military has the coordination frameworks | Draft Complete | | 3 | The PARA Experiment | Vault case study proves the principles | Draft Complete | | **4** | **The Middleman Trap** | **Adobe: what happens without doctrine** | **This Paper** | --- ## FOOTNOTES [^1]: ADBE stock data from NASDAQ real-time pricing and Yahoo Finance historical data, accessed February 28, 2026. [^2]: Goldman Sachs Sell rating ($290 target) reported by 24/7 Wall Street, February 18, 2026. Jefferies downgrade from $400 to $290 reported by IBTimes, February 2026. HSBC target cut from $388 to $302 reported by 24/7 Wall Street. [^3]: Adobe FY2025 revenue of $23.77 billion (+11% YoY) reported by Benzinga and StockAnalysis citing Adobe 10-K filing. 35% Photoshop AI usage rate from Benzinga analyst report, February 24, 2026. AI-influenced ARR exceeding $5 billion reported by Motley Fool citing Adobe Investor Day presentation. [^4]: Analyst downgrades and price target cuts compiled from Benzinga, 24/7 Wall Street, IBTimes, and StockAnalysis, February 2026. [^5]: Adobe 40-year stock price history from MacroTrends. All-time high of $688.37 on November 19, 2021. [^6]: "Seat compression" and "efficiency paradox" analysis from Trefis, "From Tailwind to Headwind: Adobe Stock's AI Reckoning," February 4, 2026. [^7]: Google Gemini image editing launch pressuring Adobe shares reported by Benzinga, February 2026. [^8]: Anthropic automation tools identified as direct Adobe competitive threat by Trefis, February 4, 2026. Reference: "The competitive landscape shifted significantly on February 3, 2026, when Anthropic introduced automation tools that pose a direct threat to Adobe's core business model." [^9]: "SaaS Apocalypse" framing from Seeking Alpha, "Adobe: The SaaS Apocalypse Is About To Meet A Valuation Check," February 2026. [^10]: Sigma level assessment based on field testing of text rendering accuracy across 10 generation attempts per model. Firefly native produced readable text in 2/10 attempts; Gemini produced readable text in 8/10 attempts. Statistical inference from small sample — formal study with larger N recommended. --- *Published via Obsidian Publish | Herding Cats in the AI Age Series* *© 2026 Jeep Marshall. All rights reserved.* ## Related - [[Index - Herding-Cats-in-the-AI-Age]] — parent folder