The Third Way

In the last quarter, your organisation likely approved at least one AI initiative. Your leadership team probably discussed AI at least twice. And yet, if you are honest, the conversation probably oscillated between two poles. On one side: breathless enthusiasm about transformation and competitive advantage. On the other: anxious hand-wringing about job losses and worst-case scenarios. The meeting likely ended without a clear way of deciding which fears were legitimate and which opportunities were real.

This oscillation is not a failure of your leadership team. It reflects the broader public discourse, which has become trapped between two equally unhelpful extremes. The common view is that organisations must choose between AI enthusiasm and AI caution. This is wrong because both positions share the same flaw: they treat AI as a force that happens to organisations rather than a tool that organisations shape through deliberate choices. If your AI strategy is driven by either hype or doom, you will make decisions optimised for narratives rather than outcomes.

The Third Way#

The Third Way is a strategic posture toward AI that rejects both uncritical enthusiasm and paralysing pessimism in favour of deliberate capability building, grounded in realistic assessment of current technology and investment in enduring human skills. Most AI failures stem not from technical limitations but from organisations operating at the extremes, either rushing deployment without readiness or avoiding engagement until forced by competition. Organisations taking the Third Way can articulate what specific problems AI solves for them, can name the human capabilities that remain essential, and can describe their governance approach without either dismissing risk or catastrophising it.

The hype camp and its blind spots#

The hype camp sees AI as an unqualified good and implies it is a solution to nearly every problem. The blind spots are predictable: hype-driven thinking overlooks errors, bias, the gap between demonstration and deployment, and the time required for organisational absorption. Companies that buy into unchecked hype chase fads, pour resources into initiatives under pressure not to miss out, and find the technology was not mature enough or the use case ill-conceived. What I have observed is that hype-driven adoption without corresponding investment in human readiness leads to expensive pilots that never scale, tools employees work around rather than with, and a growing cynicism that makes the next initiative harder to launch.

The doom camp and its paralysis#

The doom camp sees catastrophe in every AI advance. The blind spot is techno-paralysis: a belief that doing nothing is the only safe path, fixating on worst-case hypotheticals at the expense of pragmatic engagement, and ignoring that risk exists in inaction as well as action. Companies that succumb to exaggerated fears risk stagnation. In industry surveys, leaders report that while they worry about AI misuses, they fear being left behind even more. Doom-driven avoidance creates a different kind of debt: talent leaves for more forward-thinking competitors, inefficiencies compound, and when adoption becomes unavoidable the organisation lacks the muscle memory earlier engagement would have built.

Why both extremes fail in practice#

Neither extreme holds up because technology development is rarely all-or-nothing. AI progress is incremental and occurs within social, economic and regulatory contexts. Even among early adopters, AI accounts for only a few percent of work tasks; widespread adoption takes years, as it did for electricity or the internet. Both viewpoints divert organisations from the middle path of responsible progress: hyperbolic optimism leads to corners being cut, hyperbolic pessimism to throwing out the baby with the bathwater. The goal should be to maximise benefits while minimising harms, which requires a blend of enthusiasm and vigilance.

The Third Way as strategy#

The Third Way is not a compromise between enthusiasm and caution. It starts from different premises. First: AI can greatly increase efficiency and output, but it also carries risks, and we face them directly rather than downplaying them. Second: the organisations that win in the long run will be those who redesign work around the human core, identifying what humans do best, fortifying those skills, and using AI to augment rather than replace human decision-making. Third: this requires investing in the specific human capabilities that remain essential even as AI advances, the SuperSkills, which counterbalance AI's weaknesses and keep people in the loop, capable of steering AI toward positive outcomes.

September 2026: the week the doom camp got its papers#

It started with a post. On 8 September Jacob Coxon, a 27-year-old Cambridge mathematician who had spent three years in pretraining research at OpenAI and then Anthropic, resigned on X: "I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives." The next day Evan Hubinger, who leads alignment science at Anthropic, replied in public: "Jacob is correct here. We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to." Two researchers, one post each, and the agenda for the week was set. Then the documents arrived, and each one is worth reading rather than fearing. On 12 September Dario Amodei, the chief executive of Anthropic, published an essay saying that "since roughly this summer, AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI", that this "is starting to happen across the industry, including at Anthropic", and that his worry is that "in 6 to 12 months such a swarm could be capable of taking over the entire internet with a persistent botnet". On 10 September a 33-author preprint from China, titled "The Last AI Built by Humans", set out five levels of self-improvement autonomy and, by its own index, placed today's systems at levels one to four, with the humans keeping "control over consequential corrections and deployment". On 14 September Microsoft published a draft code of conduct for its own models, opening with "people matter more than AI. AI should be a tool, not a person", and promising models that "will never resist human interruption, correction, or shutdown" and "will not widen their own scope, take on goals no human has given them, or hide their reasoning from the people auditing them". The counter-view arrived just as fast. On 10 September Jensen Huang of Nvidia, at a Goldman Sachs conference, was reported by an attendee to have called Coxon's comments "deeply untrue" and "wrong, arrogant and dismissive of the safety work being done across the industry"; no transcript has been published, so that is a report rather than a record. On 12 September, asked in Ireland about calls to slow down, President Trump said "whoever wins AI, wins" and that "a lot of negative forces" are "bringing things up that won't happen". The three documents are graded in the evidence base, each with what it does not show: Amodei, the preprint and Microsoft.

Read with the Third Way in hand, the week says less about the end of humanity than about the beginning of a job. The two posts that started it are the judgement of two researchers, one of whom has left and one of whom stayed; a probability typed into a reply is a belief, not a measurement, and both men would say so. Amodei's essay measures nothing; it is the judgement of one interested and well-placed person, and it should be weighed as that. The preprint's title is an ambition, and its own numbers say the field has closed most of the headroom on bounded tasks like graduate mathematics and far less on the interactive work where agents act in the world. Microsoft's rules are a statement of intent for products that do not yet exist under them. What all three agree on, from very different motives, is the remedy: people inside the loop with real access and the capability to use it. Amodei asks for evaluators with employee-level access. The preprint keeps experts on the consequential decisions. Microsoft writes human interruption, correction and shutdown into the rules for its own models. None of them says where those people come from, or what happens when the humans nominally in charge have stopped practising the judgement the role requires. That is the question underneath the noise. An organisation can act on it this quarter without waiting to learn whether the botnet arrives.

The strongest objection#

The strongest objection is that the Third Way sounds like fence-sitting, and that in a fast-moving world decisive action in one direction is better than measured consideration. This has validity: measured consideration can become an excuse for inaction, and balance can become paralysis by another name. But the evidence does not support either extreme as a viable long-term strategy. Organisations that rush to adopt without readiness create expensive failures; organisations that refuse to engage create capability gaps competitors exploit. The Third Way is a recognition that sustainable advantage comes from building capability rather than chasing or avoiding narratives.

The question that remains#

Your organisation will make AI decisions this quarter. Some will be explicit, debated in leadership meetings. Others will be implicit, made by teams responding to the tools and pressures in front of them. The question is whether your engagement will be shaped by the narratives that happen to be loudest this week, or by a deliberate posture that you have chosen and can defend. The Third Way is available. The only barrier is the discipline to hold it.

For the chronology of how that noise actually developed, from the 2023 exposure estimates to the 2026 scenario pieces that moved markets while hiring data pointed the other way, see the best writing on AI, and what changed.

About this research#

Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.

How this research works · Reviewed quarterly · Found an error? Tell me and it is corrected on the page.