Kolt Regaskes：2027 才是 AI 冲击就业的真正转折点
一个行业观察者对 2027 就业冲击的判断：computer use 一旦可靠，端到端任务就能无人化，中间规模公司最受伤。他还劝大家学管智能体，把 AI 监督者当最后一岗。做职业规划的打工人值得看。
Kolt Regaskes 在 X 上发文判断，2026 年 AI 行业内部发展顺利，但对就业市场的实质冲击将出现在 2027 年。他认为当前模型智能水平已经足够，瓶颈在于 harness 和产品化功能仍存在缺口与低效，一旦打通，模型就能脱离演示场景、独立完成端到端任务。
他特别强调 computer use（计算机操作）能力被低估：如果变得可靠且速度够快，大量岗位将受直接影响。成本结构方面，只有非常大或非常小的公司能承担 API 和云计算开支，中间规模企业会最难受；小型公司因敏捷反而占据优势。他还担忧部分供应商的整个部门被 AI 替代，客户可能收购后只保留线下岗位。
文中指出 UBI 类支持政策几乎没有推进，也不认同“AI 会创造足够多新岗位”的说法。他给出的建议是：现在开始学习管理智能体，多数人的最后一份角色会是某种形式的 AI 监督者。
2026 has been good inside the AI bubble, but in terms of the bigger effect, I think 2027 is the actual year of the turning point.
And I don't mean that to be 100% positive either; I do feel like we'll see a much more dramatic effect on the job market.
We know these models are already excellent in terms of intelligence, but their harnesses and their features are still lacking. Gaps here, inefficiencies there. But - ignoring all these fancy demos released on launch days - it's so, so close to performing full end-to-end tasks without any human.
As I've said a few times already, I think computer use is much bigger than people are suggesting. If that becomes reliable and much faster, then people's jobs will be very much affected.
But companies need to be willing to pay for the APIs and the cloud computing. If you're very big or very small, you should be able to take the hit, but all the companies in the middle are really going to struggle.
I also feel smaller companies now have the advantage: they are much more agile, and they can change their systems much quicker. They can take gambles.
And I fear for a bunch of suppliers whose whole departments could get replaced with AI, to the point that the client could buy them out and take only the offline workers with them.
There is still very little in the way of UBI-like support yet; it is barely being discussed. And no, I don't buy this timeline of AI creating new jobs: sure, it'll create a few, but most of us will not seamlessly transfer into those roles.
For me, this is the time to start learning how to manage agents. I think for most of us, our last role will be some form of AI supervisor. Make sure that's you, help others to be the same.
本条情报归属 AITOP「行业」核心追踪栏目。在跨区域产业观察坐标中，该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线，反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括：computer use、智能体、就业、UBI，读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明：本文原始线索采编自「@koltregaskes」（@koltregaskes），经 AITOP 多维评分管线结构化提炼。
