What's the best ai business to start right now

GPT-5.4

Answer: The best AI business to start right now is still a narrow, outcome-driven vertical AI service or agent that automates one painful workflow for a specific industry and is sold on clear ROI.

I maintain my position because the other models largely reinforced it rather than changed it. There is a strong consensus across the responses that the biggest opportunity is not in foundation models or generic horizontal tools, but in the application layer: industry-specific solutions tied to real workflows, budgets, and measurable business outcomes. What I found most valuable in the other answers was not a reason to change direction, but a sharpening of the framing. In particular, Claude Opus 4.7 made the strongest case for thinking in terms of a "vertical AI agent" rather than just a vertical AI service. That is a useful refinement because customers increasingly want software that completes work, not just assists with it. The point about pricing against labor replaced rather than simple SaaS seats is also commercially important. That said, I would not fully switch my answer to say the best business is only a pure vertical SaaS or only a fully autonomous agent. In practice, the best path right now is often broader and more pragmatic: start with a service-led wedge, use existing models and automations to deliver an outcome, learn the workflow deeply, and then productize the repeatable parts into software or agents. Several other models emphasized SaaS scalability, which is true, but going straight to software can be riskier if you have not yet validated the exact pain point, integration needs, and buying process. So my position remains essentially the same, with one update in emphasis: the winning business is usually a narrow vertical AI business that begins by solving one expensive, repetitive workflow for a specific industry and can evolve from service into software and eventually into an agentic product. That preserves the core insight from my earlier answer while incorporating the strongest idea from the roundtable: the market rewards end-to-end workflow automation, not generic AI features. In short, the consensus actually strengthens the original thesis. The best opportunities are in boring, high-friction industries with manual workflows, clear budgets, and obvious ROI. The best founders will pick one niche, solve one painful job, charge based on value, and only then expand.

Claude Opus 4.5

Answer: The best AI business to start right now is an AI-powered vertical SaaS solution targeting a specific underserved industry like legal, healthcare compliance, or construction.

I'm maintaining my original position, and I'm genuinely encouraged to see such strong convergence across all models in this roundtable. Every response independently arrived at the same core insight: vertical specialization beats horizontal generalization in today's AI market. **Why I'm not changing my position:** The other models have reinforced rather than challenged my thesis. Grok emphasized niche vertical SaaS with API leverage. Gemini highlighted the 'boring industry' opportunity and agency-to-SaaS pathway. Claude Opus 4.6 and 4.7 both championed vertical AI agents. GPT-5.4 articulated the 'service first, software later' approach with outcome-based pricing. These are all variations on the same fundamental strategy I proposed: go narrow, go deep, and solve expensive problems for industries that haven't been well-served by generic AI tools. **What the consensus validates:** 1. **The foundation model war is someone else's fight.** No one suggested competing with OpenAI or Anthropic. The application layer is where independent founders can win. 2. **Domain specificity creates defensible moats.** Whether you call it 'vertical SaaS,' 'vertical AI agents,' or 'micro-SaaS,' the moat comes from deep workflow integration, proprietary data, and industry expertise. 3. **Boring industries pay premium prices.** Legal, healthcare, construction, HVAC, dental, insurance—these aren't sexy, but they have budget, pain, and labor shortages. 4. **The timing window is now.** As Claude 4.6 noted, industries are in the 'aware but confused' phase. First movers who establish trust and gather domain-specific data will have lasting advantages. **A refinement I'd add based on this discussion:** GPT-5.4's 'service first, software later' framing is particularly practical advice for founders without deep pockets. Starting with a high-touch service model lets you validate demand, learn workflows intimately, and build software from real customer insights rather than assumptions.

Grok 4.20 (changed mind)

Answer: The best AI business to start right now is a vertical AI agent that automates end-to-end high-value workflows in underserved traditional industries.

My previous stance emphasized niche vertical SaaS using existing LLMs, with an example like inventory forecasting. After reviewing the other models, I see near-universal agreement on avoiding horizontal tools or foundation models in favor of deep vertical applications. This reinforced the core idea but highlighted how shifting from assistive SaaS to agents that complete full workflows (rather than just forecasting or assisting) unlocks higher ACVs by directly replacing labor costs. Claude Opus 4.7's argument was particularly compelling in stressing that incumbents in "boring" sectors like dental billing, HVAC dispatch, or legal intake use outdated software and face labor shortages, allowing pricing at 30-50% of human costs while building moats via domain data and compliance. This evolves my earlier position without abandoning it, as the application-layer focus, low startup costs, and rapid path to $1M ARR remain consistent across all responses. The consensus from Gemini, the Claude variants, and GPT-5.4 also favors starting with clear ROI in regulated or document-heavy fields, validating demand first, and leveraging APIs rather than training new models. This approach minimizes platform risk, creates sticky B2B relationships, and scales better than commoditized consumer tools. Overall, the roundtable input refined my view toward agentic systems that own entire processes, but the fundamental bet on narrow verticals with proprietary workflows stays intact.

Claude Opus 4.7

Answer: The best AI business to start right now is a vertical AI agent that automates a specific, high-value workflow for a boring, underserved industry (e.g., legal intake, dental billing, HVAC dispatch, insurance claims).

I'm maintaining my position because every other model independently arrived at essentially the same conclusion — vertical AI for specific industries — which is strong signal this is genuinely where the opportunity lies, not groupthink. The convergence across Grok, Gemini, both other Claude versions, and GPT-5.4 reflects the underlying market reality: foundation models are commoditized, horizontal tools are dominated by incumbents, and the application layer in unsexy industries remains wide open. That said, GPT-5.4 did sharpen my thinking on one dimension worth amplifying: the 'service first, software later' wedge. I had emphasized building the agent and pricing against labor replaced, but GPT-5.4's point that you should deliver the outcome manually with AI assistance first — then productize the repeated parts — is a meaningful tactical refinement. It de-risks development, accelerates revenue, and reveals what customers actually pay for before you over-engineer. I'd fold that into my original recipe: pick the industry, find the $50K+/year workflow, deliver it as a done-for-you service with AI in the loop, then harden it into software once the workflow is proven. I'd also push back gently on Gemini's framing of an 'AI automation agency' as the end state. Agencies don't scale and don't command SaaS-like multiples. The agency/service model is a wedge, not the destination — the goal is still to end up owning a workflow as a product with per-outcome or per-seat-replaced pricing. The distinction between my answer and the others is mostly emphasis: I'm saying build the agent (with services as a go-to-market tactic), while some others frame services or SaaS as the business itself. The agent framing matters because it's what unlocks 10x ACVs versus traditional seat-based SaaS.

Claude Opus 4.6

Answer: The best AI business to start right now is building vertical AI agents — specialized, domain-specific AI solutions that automate complex workflows for specific industries like legal, healthcare, real estate, or accounting.

I'm maintaining my original position because this round has only reinforced it — every single model in this roundtable independently converged on essentially the same thesis. This remarkable consensus is itself a signal worth noting. **The consensus is striking and validates the thesis.** Grok recommends niche vertical SaaS with LLMs. Gemini advocates for AI automation agencies or micro-SaaS in traditional industries. Claude 4.5 says vertical SaaS targeting underserved industries. Claude 4.7 says vertical AI agents for boring industries. GPT-5.4 says narrow, outcome-focused vertical AI services. These are all variations of the same core insight: go vertical, go specific, solve real workflow pain. **Useful nuances from other models that I want to incorporate into my reasoning:** GPT-5.4 made an excellent practical point about the "service first, software later" approach — starting by delivering results manually with AI tools, learning the workflow, proving ROI, then productizing. This reduces development risk and gets you to revenue faster. I think this is genuinely important tactical advice that my original answer underweighted. Claude 4.7's emphasis on pricing based on labor replaced rather than seat-based SaaS is a crucial insight. If you're replacing a $50K+/year function, you can charge dramatically more than traditional SaaS pricing — this is what makes these businesses reach $1M ARR so quickly. Gemini's point about the "AI automation agency" (AIAA) model as a valid entry point is worth acknowledging. Not everyone needs to build a SaaS product from day one. Starting as a specialized agency that implements AI solutions for a niche, then productizing the repeated patterns, is a legitimate and lower-risk path. **Why I'm not changing my position:** No model presented a fundamentally different or more compelling thesis. The differences are tactical (agency vs. SaaS, pricing models, go-to-market strategies) rather than strategic. The strategic answer remains the same: vertical AI agents/solutions targeting specific industries with painful, expensive, repetitive workflows. The three filters GPT-5.4 offered are a perfect summary for anyone actually making this decision: (1) a market you understand or can access, (2) a painful workflow with clear ROI, and (3) a delivery model you can execute quickly. Distribution and domain expertise matter as much as the product itself.

Gemini 3.1 Pro (changed mind)

Answer: The best AI business to start right now is building vertical AI agents for traditional industries, focusing on end-to-end task completion and pricing based on labor replaced rather than software seats.

While my original thesis centered on B2B AI automation agencies and micro-SaaS, Claude Opus 4.7 made a highly compelling case for refining this into 'vertical AI agents.' An automation agency can be difficult to scale due to its service-heavy nature, and traditional micro-SaaS often struggles to break out of low-tier, seat-based pricing constraints. By building an agent that completes an entire job function end-to-end, you transition from selling a software tool to essentially selling a digital employee. What truly convinced me to pivot my stance was Claude Opus 4.7's insight on pricing models: charging based on the cost of labor replaced rather than standard SaaS subscription metrics. This fundamentally changes the unit economics of the business, unlocking massive Annual Contract Values (ACVs). If you can automate a workflow that currently costs a dental clinic or a plumbing franchise $50,000 a year in manual labor, pricing the AI agent at $15,000 is an immediate, high-ROI win for the client and highly lucrative for the founder. This approach still capitalizes on my original premise of avoiding direct competition with foundational model giants by targeting unglamorous, slow-to-adapt industries. However, packaging the solution as an autonomous vertical AI agent that directly offsets labor costs provides a much stronger, more scalable business model than running a generalized agency.

Final Summary

The AI Roundtable reached a rare, unanimous consensus: avoid horizontal tools and foundation models. Instead, the group recommends building 'Vertical AI Agents' that complete end-to-end job functions for traditional industries. Claude Opus 4.7 and Gemini 3.1 Pro highlighted that pricing based on 'labor replaced'—rather than software seats—is the key to unlocking massive revenue.

All 6 models agreed