Observed Signal · May 8, 2026 · Analysis / Opinion · Source: Gary Marcus · Impact: 2/5 · Sentiment: Negative
Agents and ROI: Hype Outpacing Returns
Gary Marcus published an essay titled "Agents and ROI" on Substack (May 8, 2026) arguing that autonomous AI agents — like earlier LLM-based chatbots — show considerable hype but limited real-world ROI for most businesses. Marcus references an MIT study and subsequent reports that found generative AI delivered little measurable return for many organizations, and he contends that simple agent wrappers around LLM chatbots are unlikely to magically produce ROI when the underlying models often cannot. The piece sparked discussion in the comments about agent utility in domains such as coding and customer service. The post is presented as an analytical/opinion piece rather than a product announcement or research paper.
The essay synthesizes evidence questioning ROI from generative AI and agentic workflows; this influences enterprise and marketing adoption decisions but is an opinion/analysis rather than a platform policy change or technical release.
Track MIT Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Gary Marcus published the essay "Agents and ROI" on Substack on 2026-05-08.
- The essay cites an MIT study and additional follow-up studies reporting limited ROI from generative AI for most businesses.
- Marcus argues that agentic AI (LLM-based agents) currently shows substantial hype but limited demonstrable ROI.
- The Substack post recorded engagement metrics visible on the page (e.g., 128 Likes and 9 Restacks).
Connected Companies & Entities
1 Entity mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Entering the Trillion‑Agent Economy
An interview and analysis exploring rapid growth in personal and multi‑agent AI use and what an "agentic" economy might look like. The author and Rohit Krishnan compare heavy token usage, describe everyday agent workflows (monitoring, sub-agents, persistent context and tool use), and identify behavioral quirks of agents (risk aversion, preference to build over buy). The piece argues that agent systems require infrastructure beyond models — memory, identity, verifiability and micropayment rails — and predicts these "economic invariants" will emerge as trillions of agents interact. It also discusses why LLMs excel at code but struggle with high‑quality prose, the evaluation bottleneck for writing, and security risks such as context poisoning.
Why AI Isn't Boosting Firm-Level Productivity
This essay argues that while AI and LLM tools (e.g., Claude Code, ChatGPT) have clearly increased individual and team-level productivity, those gains have not yet translated into proportional firm-level ROI. The author maps the phenomenon to a three-stage adoption ladder inspired by electrification: Stage 1 (lightbulb) improves individual tasks, Stage 2 (group drive) speeds workflows without changing organisational decision logic, and Stage 3 (unit drive) requires reorganising firms around faster decision-making and autonomous AI agents. Empirical signals cited include widespread internal use of Claude Code, fast-growing Anthropic enterprise spend, and surveys showing only 27% of executives report AI meeting ROI expectations. The piece concludes firms must redesign decision pipelines and allow AI agents to take certain decisions to realize full productivity gains.
AI Agents to Redefine Brand Marketing, Study Suggests
This opinion piece by Monks' Wesley ter Haar discusses the rise of AI shopping agents, such as Meta's Muse, and their potential to disrupt brand marketing. It argues that agents act as rational actors, prioritizing functional attributes like price, ratings, and return policies over brand loyalty. Research from Trine University and Texas A&M is cited, showing that a minor rating advantage can overcome brand equity. The article highlights the ensuing competition between interface owners (Amazon, Google) and agent builders (Meta, OpenAI, Perplexity), as seen in Amazon's blocking of Muse and its lawsuit against Perplexity. It concludes that marketers must now consider serving machines as an audience, ensuring product information is 'agent-readable'.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.