TJE Ventures

A Caution About Predictions

AI forecasting has a poor track record. Predictions made in 2020 about where AI would be in 2023 were simultaneously too conservative on capability and too optimistic on deployment. The technology surprised nearly everyone. Acknowledging this upfront matters: what follows is a considered view of likely directions based on current trajectories, not certainty. Build strategy around the directions, not the specific claims.

What Is Highly Likely

Agentic AI will become mainstream

The shift from AI as a responder to AI as an actor --taking sequences of actions to complete multi-step goals --is already underway. In the next two to three years, AI agents are likely to become as ordinary in business software as chatbots are today. Automated workflows, AI-managed email triage, agents that coordinate across systems to complete tasks end-to-end --these will move from experimental to routine. The organizations building familiarity with agentic concepts now will have a significant head start.

Multimodal AI will be the default

Today’s most capable AI models already handle text, images, audio, and increasingly video within a single system. This multimodal capability is becoming the default rather than the exception. In practical terms, this means AI that can analyze a photograph of a document, describe what is in a video, interpret a chart, and respond to voice --all within the same interaction. Business applications that currently use separate specialized tools will consolidate into unified AI interfaces.

Costs will continue to fall significantly

The cost of AI inference --the per-query cost of running a model --has dropped by orders of magnitude over the past three years and is expected to continue falling. What costs dollars today will cost cents. What costs cents today will become effectively free at scale. This has significant implications for which use cases become economically viable and which AI deployment models make sense for different organizations.

AI will be embedded everywhere

The “AI product” as a distinct category will become less meaningful as AI capabilities are embedded into every software product your organization already uses. CRMs, ERPs, project management tools, communication platforms, HR systems --all of these will have AI deeply integrated. The skill that matters is not knowing how to use a specific AI tool; it is knowing how to work effectively with AI-augmented systems in general.

Regulatory frameworks will mature

The EU AI Act is already in force and implementing in stages through 2027. US federal and state regulation is developing unevenly but directionally toward more oversight. Industry-specific regulation in financial services, healthcare, and other sectors is tightening. Organizations that have built governance frameworks early will adapt more easily than those that face compliance as a surprise.

What Is Plausible But Uncertain

Significant disruption to knowledge work roles

The labor market impact of AI on knowledge work is uncertain in timing and distribution, but the direction is clear: AI will automate more tasks, shift role compositions, and create some new categories of work while reducing demand for others. The pace of this shift is the key uncertainty. Conservative estimates suggest gradual adjustment over a decade; more aggressive forecasts see meaningful disruption within three to five years. Preparing teams now --as covered in the Preparing Your Team for AI article --is the hedge that works regardless of the pace.

AI reasoning and reliability improving substantially

Current AI systems hallucinate, make logical errors, and struggle with multi-step reasoning in ways that limit how much autonomy they can safely be given. Research on improving AI reliability --through better training, verification techniques, and architectural approaches --is active and showing results. If this trend continues at its current rate, the AI of 2027 may be reliable enough to be trusted with significantly more autonomous action in high-stakes contexts. This would change the risk calculus for many AI governance decisions.

Commoditization of frontier AI capability

The performance gap between the best available AI models and the open-source alternatives has been narrowing. If this continues, frontier-quality AI capability may become available to self-host within two to three years, eliminating the cost and data privacy trade-offs that currently drive many organizations toward cloud AI. This would particularly benefit organizations with strict data sovereignty requirements.

What Businesses Should Be Preparing For Now

Build AI literacy at every level

The organizations that benefit most from the next wave of AI capability will be those where employees at all levels understand how to work with AI effectively. This literacy compounds: teams that have been using AI for two years will get more from more capable future AI than teams just starting. Start building this capability now.

Develop governance before you need it

Regulatory requirements will only become more demanding. Organizations that have documented AI governance frameworks, accountability structures, and audit trails will be better positioned to demonstrate compliance when required. Building governance reactively --in response to a regulatory deadline or an incident --is more expensive and less effective than building it proactively.

Stay oriented without chasing every development

The AI landscape produces significant announcements weekly. Most do not require immediate action. A useful discipline: assign someone in your organization to monitor AI developments and filter for what is actually relevant to your business. Quarterly briefings on significant developments are more valuable than daily noise. The goal is informed readiness, not exhausting vigilance.

Think in terms of capability, not tools

Specific AI tools will change. The underlying capability --AI that can read, write, reason, and act --will persist and grow. Investing in organizational ability to use AI effectively, rather than expertise in specific tools, is the more durable bet.

“The organizations that will benefit most from AI in three years are not necessarily those deploying the most AI today --they are those building the understanding, the judgment, and the structures to use AI well as it continues to evolve.”

← Previous Next: You Cannot Run a Ferrari on Kerosene →