In late 2025, “agentic AI” stopped being a slideware buzzword and started to look like infrastructure.
Within a few weeks, three very different players—Target in retail, Ardent Health in healthcare, and Adobe in marketing tech—each announced major deployments of AI agents that don’t just answer questions but do work: filling baskets, writing medical notes, optimizing campaigns.
These aren’t experimental demos. They’re production systems, wired into logistics networks, clinical workflows, and revenue pipelines.
This essay looks at those three moves as a single story: how agentic AI is becoming the backbone of new business workflows, what that means for labor, trust, and regulation, and how it hints at a post-labor economy where “apps” quietly turn into “agents”.
1. Setting the Stage: From Chatbots to Agents
Classic chatbots were reactive: you type, they reply. Agentic AI adds three crucial capabilities:
- Goal orientation – you express an outcome (“plan my holiday shopping”, “document this encounter”, “improve pipeline on this account segment”), and the system decomposes it into steps.
- Tool use and orchestration – the AI triggers actions in external systems (carts, EHRs, marketing platforms, scheduling, analytics).
- Memory and feedback loops – it learns from prior outcomes, improving recommendations or workflows over time.
OpenAI recently disclosed that more than 1 million businesses now use its products—either via ChatGPT for Work or via its developer platform—making it the fastest-growing business software platform in history. OpenAI That’s the substrate on which Target’s new shopping experience is being built. Meanwhile, vertical platforms like Ambience Healthcare and category giants like Adobe are layering their own specialized agents on top.
What’s new in late 2025 is not just that AI can talk; it’s that AI is increasingly being trusted to act inside core workflows.
2. Retail: Target and the Rise of Agentic Commerce
2.1 What Target and OpenAI Actually Announced
On 19 November 2025, Target announced a “first-of-its-kind conversational, curated shopping experience” inside ChatGPT. Target Corporation This isn’t a mere product search plugin—it’s effectively a full Target shopping front-end embedded in a conversational agent.
Key capabilities of the Target app in ChatGPT:
- Curated browsing and ideas: the assistant proposes themed lists (“host a cozy Friendsgiving”, “STEM gifts for a 10-year-old”) instead of forcing users to craft detailed filters and searches. Target Corporation+1
- Multi-item baskets in a single transaction: users can add multiple items suggested in conversation and check out once, not item-by-item. Target Corporation+1
- Fresh food and full category coverage: not just durable goods, but groceries and perishables. Digital Commerce 360+1
- Omnichannel fulfillment: Drive Up, Order Pickup, and shipping; same-day delivery is on the roadmap. OpenAI+1
- Target Circle integration: loyalty benefits and personalized offers carry over into the ChatGPT environment. Brand Vision
In parallel, Target is deepening its internal AI use. It’s rolling out ChatGPT Enterprise to about 18,000 headquarters employees for tasks such as supply-chain forecasting, store process optimization, internal support, and AI-driven customer tools. OpenAI+1
So this is not a single “bot project”—it’s a multi-layer partnership:
- Consumer-facing agent for shopping
- Employee-facing agents for knowledge work and operations
- Model-level integration into Target’s digital products and internal tools
2.2 Why This Is Different From “Just Another Channel”
From a systems perspective, Target is doing something quite radical: it’s allowing a non-Target interface (ChatGPT) to become a first-class storefront.
Historically, e-commerce controlled the entire presentation stack: the app, the website, and the search and recommendations within them. Here, Target exposes:
- Product catalog & availability
- Pricing and promotions
- Cart creation & transaction APIs
- Fulfillment options
…to an external agent that can decide how to present, combine, and sequence those items according to a user’s conversational goals.
That is “agentic commerce” in practice:
- The user states a goal (“help me plan a Marvel-themed birthday party for 8-year-olds under $150”).
- The agent searches catalog, applies constraints, proposes a list, and iterates as the user refines.
- The agent builds and updates the cart, evaluates trade-offs (substitutions, out-of-stock), and executes checkout.
From Target’s perspective, this could shift key metrics:
- Higher conversion for complex missions (holiday planning, back-to-school, events) where a conversational planner outperforms manual browsing.
- Higher average order value (AOV) because the agent can remind you of forgotten complements (napkins with plates, batteries with toys, etc.).
- Deeper personalization as the model learns shopper patterns across sessions and missions.
But it creates equally serious challenges.
2.3 Risks and Governance in Agentic Commerce
Allowing an AI agent to drive carts and checkouts raises new trust questions:
- Hallucinated or misleading suggestions: if ChatGPT suggests a “gluten-free” item that isn’t actually labeled gluten-free, who is responsible?
- Fairness of recommendations: will shoppers consistently be steered toward higher-margin products? Will this be regulated as a form of “algorithmic steering”?
- Data sharing and privacy: user intents, conversations, and cart behavior are now co-processed by two companies (OpenAI and Target), which must align their privacy, retention, and consent policies.
Technically, Target mitigates risk by tightly controlling:
- The product and inventory data exposed to the agent
- The checkout flow, which relies on the existing Target account and payments stack
- The policy constraints, such as age-restricted items, returns, and fulfillment limits
But the strategic shift is clear: Target is betting that interface power will increasingly move to agentic layers like ChatGPT, and that being a first-mover store inside that agentic layer will be worth the risk.
3. Healthcare: Ardent Health and Ambience’s Ambient AI Scribe
If Target shows how agents can act on carts, Ardent Health shows how they can act on clinical documentation, one of the highest-stakes workflows in any industry.
3.1 The Ardent–Ambience Rollout
On 17 September 2025, Ardent Health announced an enterprise-wide rollout of Ambience Healthcare’s AI platform across its ambulatory network. ardenthealth.com+1
Key context:
- Ardent operates 30 hospitals and ~280 sites of care across six U.S. states. Fierce Healthcare
- Ambience’s platform covers AI scribing, coding, and clinical workflow support, integrated with systems like Epic. ardenthealth.com+1
- The rollout follows a pilot spanning 17 specialties and 7 languages, with clinician utilization rates around 90%. ambiencehealthcare.com
Measured outcomes from the pilot and early deployments include:
- ~45% decrease in documentation time, according to Epic UAL data. ardenthealth.com+1
- ~5 hours per week saved per clinician on documentation. Fierce Healthcare+1
- Reported reductions in cognitive load and improved job satisfaction, with high percentages of clinicians agreeing that the tool made their day more sustainable. ambiencehealthcare.com
These are substantial numbers. For a large health system, five hours per week per clinician, multiplied across hundreds or thousands of physicians and advanced practice providers, translates into tens of thousands of recovered hours annually.
3.2 How Ambient Agents Work in the Exam Room
An “ambient AI scribe” is a specialized agent built around a fairly standard pipeline, but tuned to clinical constraints:
- Capture: it listens to the patient–clinician conversation (with consent), often via a smartphone or room microphone.
- Transcription & structuring: it converts speech to text and segments the dialogue into problems, history, exam, assessment, and plan.
- Documentation synthesis: it produces a draft clinical note aligned with the health system’s templates and the requirements of billing and coding.
- Coding and quality checks: it proposes billing codes, flags potential compliance issues, and embeds structured data for quality reporting.
- Clinician review and sign-off: the human remains the final arbiter; they edit, approve, and sign.
What makes it agentic is that the system:
- Adapts to clinician preferences (“I prefer bullet-point plans”, “don’t restate labs I’ve just reviewed”).
- Interacts with the EHR to pull relevant prior data or problem lists.
- Optimizes its own prompts and templates based on feedback and usage patterns.
3.3 The Human Impact: From Burnout to Redesign
Healthcare is one of the sectors facing intense burnout, driven in part by “pajama time”—hours of charting after formal clinic is done. A 45% reduction in documentation time is not just an efficiency statistic; it’s a potential workforce stabilization tool. Fierce Healthcare+1
But there’s a deeper post-labor implication:
- Today, ambient AI is framed as “giving clinicians time back”.
- Tomorrow, it could be framed as “redefining the role of the clinician entirely”—shifting cognitive load from note-taking and billing compliance to shared decision-making and patient relationships.
That redesign, however, depends on trust:
- Clinical accuracy: speculation or hallucination in a note is not just annoying; it can be dangerous.
- Attribution & provenance: who is responsible if a hallucinated statement ends up in the record and influences downstream care?
- Data protection: PHI is among the most regulated data types in the world. Any agent touching it must live inside strict governance boundaries.
Ambience and similar vendors are responding with:
- Strong on-prem or VPC deployment options and tight EHR integrations.
- Guardrails around off-label summarization (e.g., avoiding guessing missing details).
- UI patterns that make AI-generated content highly visible and easily editable.
Still, the Ardent rollout marks a line: a major U.S. health system is betting that ambient, always-on AI agents in the exam room will be the norm, not the exception.
4. Marketing & CX: Adobe’s AI Agents and the Agent Orchestrator
If Target is about buying and Ardent is about documenting, Adobe is about orchestrating entire customer journeys with agents.
4.1 GA of AI Agents and Agent Orchestrator
On 10 September 2025, Adobe announced the general availability of AI agents as part of Adobe Experience Platform (AEP), along with a component called Agent Orchestrator. Adobe Newsroom+1
The first wave includes six out-of-the-box agents embedded into core AEP applications such as:
- Real-Time Customer Data Platform
- Journey Optimizer
- Experience Manager
- Customer Journey Analytics SiliconANGLE
These agents tackle tasks including:
- Audience segmentation and refinement
- Journey design and optimization
- Content recommendations and site personalization
- Insights extraction and anomaly detection in customer data Adobe Newsroom+1
The Agent Orchestrator coordinates multi-step workflows involving multiple agents and data sources, enabling reasoned, end-to-end scenarios (e.g., detect churn risk → generate a win-back journey → launch across email and paid media). CIO+1
Adobe also previewed Agent Composer, a forthcoming tool for customizing agents based on brand policies, governance rules, and workflow contexts—essentially a “no-/low-code” layer for tailoring agent behavior. CIO+1
4.2 B2B Agents and Human–AI Collaboration
Adobe is explicitly positioning many of these agents for B2B marketing and sales use cases, not just B2C. In B2B, cycles are longer, stakeholders are many, and data is scattered.
Adobe’s B2B agents focus on:
- Identifying key decision-makers and buying groups
- Scoring and prioritizing accounts
- Orchestrating cross-channel engagement for complex journeys
- Surfacing insights for revenue teams in a collaborative, human-in-the-loop way Adobe Business+1
Cisco, for example, has publicly cited Adobe’s AI agents as an “unlock” that reduces the time to identify decision-makers and orchestrate journeys, boosting engagement and accelerating deals. Adobe Business
What’s notable here is that the agents are embedded directly into existing workflows—marketers still log into Experience Cloud, but instead of manually crafting segments or journeys, they:
- State goals (“increase pipeline from mid-market accounts that have engaged with webinars in the last 60 days”).
- Let agents propose segments, paths, and content variants.
- Adjust, approve, and monitor performance, rather than building from scratch.
Where Target opens its APIs to an external agent, Adobe pulls agentic capabilities inside its own platform and exposes them as “smart co-workers” for marketers and sellers.
5. Common Architecture: What All Three Moves Share
Retail, healthcare, and marketing are wildly different domains, but these three deployments share some structural patterns that are worth calling out.
5.1 The Agentic Pattern
Across all three, you see the same high-level workflow:
- Intent capture
- Target: “Help me plan a holiday party for 10 people under $200.”
- Ardent: a live clinical encounter, implicitly defining the documentation and coding tasks.
- Adobe: “Improve conversion for these segments” or “Design a journey to reactivate dormant accounts.”
- Decomposition into tasks
- Break “plan a party” into menu, decor, activities, etc.
- Break “document visit” into HPI, ROS, exam, assessment, plan, codes.
- Break “improve conversion” into segmentation, experimentation, channel selection.
- Tool invocation and environment actions
- Target: catalog search, pricing and inventory, cart API, fulfillment API.
- Ardent: speech-to-text, EHR queries, template engines, coding engines.
- Adobe: query customer data, build segments, configure journeys, push to channels.
- Human-in-the-loop oversight
- Shopper can adjust carts and ask “why this?”
- Clinician can edit and override notes and codes.
- Marketer can accept or reject agent-proposed journeys and content.
- Learning and optimization
- Performance and usage feedback feed into better prompts, ranking, and defaults over time.
This is the “agentic loop” becoming standard: observe → plan → act → learn.
5.2 Platform Strategies and Lock-In
Each player is also clearly thinking in platform terms:
- OpenAI & Target: ChatGPT becomes a “super interface” for commerce; Target becomes one of the default retail agents living inside it.
- Ambience: focuses narrowly on healthcare, but aims to be the ambient layer across multiple health systems, languages, and specialties.
- Adobe: turns Experience Platform into an agent execution environment rather than just a data and workflow hub.
Once these agents are wired into core workflows:
- Switching costs rise (you’re not just migrating data, but agent behaviors tuned to your org).
- New ecosystems emerge (e.g., specialized prompts, domain ontologies, and guardrails as configuration products).
- “Agent templates” become proprietary assets (a best-performing collections agent, a high-conversion onboarding journey, a specialty-specific clinical note style).
6. Labor, Trust, and the Post-Labor Angle
From a post-labor economics perspective, these deployments are still in the “augmentation” phase, but they’re inching toward “substitution of tasks” for certain roles.
6.1 What Gets Automated First
Across the three domains, notice which tasks are being automated first:
- Retail: product discovery, cross-sell, basket building, channel selection—things human sales associates or merchandisers historically did, especially in higher-touch environments.
- Healthcare: transcription, structuring, and coding—highly repetitive tasks that emerged partly as a side-effect of digitization and billing complexity.
- Marketing: segmentation, targeting, journey mapping, and content personalization—cognitive work that sits between analytics and execution.
The pattern: highly structured, high-volume, cognitively demanding but low-creativity tasks are prime candidates for agentic automation.
6.2 What Humans Still Anchor
At the same time, each deployment keeps humans as the anchor point for:
- Value judgments: clinicians decide the final content of notes; marketers define goals and guardrails; shoppers decide what “good enough” feels like.
- Relationship work: doctor–patient empathy, sales conversations, brand voice nuance.
- Accountability: regulatory and legal responsibility still falls on the human or the institution, not the agent.
Post-labor economics doesn’t assume humans vanish. It assumes the composition of human work changes, with much higher leverage and more complex responsibility per worker.
In that light:
- A clinician with an ambient scribe can see more patients or spend more time per patient.
- A small marketing team can operate with the capabilities of what used to be a much larger org.
- A single shopper can orchestrate complex purchases that used to require multiple interactions with staff.
The key open question is who captures the surplus: the workers (through less burnout, more pay, or shorter weeks), the firms (through higher margins), or the platforms.
7. Governance: Safety, Regulation, and Standards
As agentic AI moves from “help me write an email” to “help me run my store / hospital / marketing engine”, governance becomes existential.
7.1 Sector-Specific Constraints
- Retail: consumer-protection rules, marketing regulations, and potential antitrust scrutiny if a few agentic platforms become dominant gateways to commerce.
- Healthcare: HIPAA, GDPR (for EU operations), and sectoral oversight on medical device software if certain AI tools influence clinical decision-making.
- Marketing: evolving privacy regimes (e.g., cookie deprecation, data-minimization rules), transparency around personalized targeting, and algorithmic fairness.
You can already see early responses:
- Adobe talks extensively about policy controls and brand guardrails embedded in Agent Composer. CIO+1
- Healthcare vendors emphasize compliance-grade deployment models and detailed audit logs for how notes and codes were generated. ardenthealth.com+1
- OpenAI and Target communicate their partnership in terms of enhancing guest experience and privacy-preserving personalization, clearly aware that regulators will be watching how agentic commerce evolves. OpenAI+1
7.2 The Coming Standards
Over the next few years, expect to see emerging standards around:
- Agent transparency – clearly indicating when an agent is acting on your behalf, what constraints it operates under, and how to override it.
- Outcome-based evaluation – moving beyond model benchmarks to domain-specific metrics (e.g., error rates in notes, adverse event correlation, mis-targeting rates in marketing campaigns).
- Interoperability – how proprietary agents (Target’s, Ambience’s, Adobe’s) interact with general platforms (OpenAI, cloud providers) without locking users into opaque ecosystems.
These standards will likely be driven by a mix of regulators, industry bodies, and large enterprise customers insisting on auditable, portable agent behaviors.
8. Looking Toward 2026–2028: What Comes Next
The Target, Ardent, and Adobe moves feel like early markers of a shift that will accelerate through 2026–2028.
Here are some grounded expectations:
- From pilots to “default workflow”
- In healthcare, ambient scribing may move from “opt-in” to default, with manual documentation becoming exceptional.
- In marketing, agent-generated journeys and content will be the baseline; completely hand-built journeys will be reserved for special cases.
- In commerce, conversational planning and buying will become a parallel default alongside traditional search and browsing.
- Bundled agent ecosystems
- More enterprises will adopt a “few big platforms + several vertical agents” strategy: e.g., OpenAI/Anthropic + industry-specific agents like Ambience + domain suites like Adobe.
- Enterprises will demand consistent policy and security layers across these agents.
- Task-level automation metrics
- Instead of “X% of jobs automated,” we’ll see more nuanced stats: “Y% of documentation steps automated”, “Z% of segmentation and campaign design delegated to agents”.
- Firms will start optimizing portfolios of tasks across humans and agents, much like they once optimized portfolios of outsourced processes.
- New roles and skills
- “Agent operations”, “workflow graph designers”, and “prompt/guardrail engineers” will become standard roles.
- Clinicians, marketers, and merchandisers will be expected to understand how to supervise agents, not just use apps.
- Regulatory inflection points
- A high-profile incident—say, an AI-driven clinical note contributing to an adverse event, or a commerce agent steering vulnerable consumers unfairly—will likely prompt new oversight.
- At the same time, positive macro-level data (reduced burnout, increased access, better personalization) will push against overly restrictive bans.
9. Conclusion: Agents as the New Runtime for Work
Taken together, the Target–OpenAI partnership, Ardent’s Ambience rollout, and Adobe’s AEP agents are early glimpses of a world where:
- Interfaces are conversational and goal-driven, not form- and menu-driven.
- Workflows are executed by agents that can reason, sequence tools, and improve over time.
- Humans become supervisors, editors, and relationship owners, rather than primary executors of repetitive tasks.
We are not yet in a fully post-labor economy, but we are very clearly in a post-app economy, where the real “program” is the agent, and apps are just tools in its toolbox.
In 2010, every company needed “a website.”
In 2020, every company needed “an app.”
By 2028, every serious organization will need agents—and a clear answer to a simple question:
What do you want your AI to do for you, when you’re not looking?
