跳到主要內容

AI Agent自建軟體評測:McKinsey曝32%企業不買SaaS | Agentic Coding Review: McKinsey Finds 32% Skip Buying SaaS

By Kit 小克 | AI Tool Observer | 2026-09-14

🇹🇼 AI Agent自建軟體評測:McKinsey曝32%企業不買SaaS

AI Agent 正在改寫企業軟體採購的規則。根據 McKinsey 最新發布的《State of AI in 2026》報告,調查全球 97 國、1,719 位企業主管後發現,32% 的公司放棄採購現成軟體,改用 AI Agent 自己寫。這不是單一案例,而是「Build vs Buy」天平正式倒向自建的訊號,也是這波 agentic coding(代理式編碼)工具成熟後最實際的商業影響。

什麼是 Build vs Buy 大逆轉?

過去企業要導入一套系統,多半直接買 SaaS——省時間、省人力。但現在 Claude Code、Cursor、GitHub Copilot 這類工具讓工程團隊能在幾天內生出堪用的內部系統,「自己寫」的成本大幅下降,於是不少公司乾脆把採購清單砍掉重練。

McKinsey 報告怎麼說?32% 企業不買軟體

這份調查在 2026 年 5 月 4 日到 6 月 8 日蒐集意見,8 月 25 日公布。關鍵數字:

  • 32% 的受訪企業因為 AI Agent 能自建,取消了原本要買的軟體
  • 在把 AI 貢獻算進至少 5% EBIT 的「高績效企業」(佔樣本 6%)中,將近一半(約 49%)選擇自建
  • 一般企業則只有 31% 這麼做——落差顯示 AI 用得越深,越敢自己動手

哪些產業改用 AI Agent 自己寫軟體?

產業分佈上,科技業比例最高,達 41%;緊追在後的是醫療保險與醫療服務業(39%),再來是專業服務與能源業(38%)。這些都是內部系統複雜、客製化需求高、又養得起工程團隊的產業,AI Agent 的邊際效益自然最明顯。

自建軟體真的划算嗎?Kit 小克的老實話

說老實話,我不會叫大家馬上把 SaaS 訂閱全砍。AI Agent 寫程式很快,但寫完之後呢?誰負責維護、修 bug、應付資安稽核?這些報告數字沒講的部分才是重點:

  • 優點:客製化程度高、不被供應商綁死、迭代速度快,尤其適合內部工具或利基流程
  • 風險:AI 生成的程式碼堆久了一樣會變技術債,而且沒有 SaaS 廠商的 SLA 和支援團隊幫你擦屁股
  • 建議做法:核心、需要長期穩定支援的系統(財務、HR)還是買現成的;內部效率工具、一次性流程,才是 AI Agent 自建的甜蜜點

這波 Build vs Buy 逆轉是真的,但它獎勵的是「有工程能力善用 AI Agent」的公司,不是每間公司都適合跟風。

好不好用,試了才知道。


🇺🇸 Agentic Coding Review: McKinsey Finds 32% Skip Buying SaaS

AI Agent tools are quietly rewriting enterprise software procurement. McKinsey's newly published "State of AI in 2026" report surveyed 1,719 business leaders across 97 countries and found that 32% of organizations skipped buying off-the-shelf software because agentic coding tools let them build it in-house instead. This isn't a fluke — it's a real signal that the Build vs Buy scale has tipped toward building, and it's the most concrete business impact of the maturing agentic coding wave so far.

What Is the Build vs Buy Reversal?

Companies used to default to buying SaaS to save time and headcount. Now, tools like Claude Code, Cursor, and GitHub Copilot let engineering teams spin up a usable internal system in days, not months. As the cost of "just building it" drops, more companies are cutting line items off their procurement list entirely.

What Does McKinsey's Report Actually Say?

Fielded between May 4 and June 8, 2026, and published August 25, the survey found some striking numbers:

  • 32% of respondents canceled a planned software purchase because AI Agents could build the equivalent in-house
  • Among "high performers" — the 6% who attribute at least 5% of EBIT to AI — nearly half (about 49%) skipped buying
  • Among everyone else, only 31% did — meaning deeper AI adoption correlates directly with confidence to build

Which Industries Are Building Instead of Buying?

The tech sector leads at 41%, followed by healthcare payers and providers at 39%, then professional services and energy at 38%. These are industries with complex internal systems, high customization needs, and engineering teams already on payroll — exactly where AI Agent's marginal cost advantage shows up first.

Is Self-Building Actually Worth It? Kit's Honest Take

I'm not telling anyone to cancel their SaaS subscriptions tomorrow. AI Agents write code fast — but what happens after launch? Who maintains it, patches bugs, and survives a security audit? That's the part the survey numbers skip:

  • Upside: deep customization, no vendor lock-in, fast iteration — great for internal tools and niche workflows
  • Risk: AI-generated code accumulates tech debt just like human-written code, minus the vendor's SLA and support team
  • My recommendation: keep buying for systems that need long-term stability and support (finance, HR); let AI Agents build the internal efficiency tools and one-off workflows

The Build vs Buy shift is real, but it rewards companies with real engineering muscle to wield AI Agents well — it's not a trend every company should chase blindly.

You won't know until you try it.

Sources / 資料來源

常見問題 FAQ

McKinsey的32%數字是怎麼算出來的?

McKinsey在2026年5月4日至6月8日調查全球97國、1,719位企業主管,發現32%的受訪企業因為AI Agent能自建軟體,取消了原本的採購計畫。

哪些公司最愛用AI Agent自己寫軟體?

把AI貢獻算進至少5% EBIT的「高績效企業」最積極,約49%選擇自建;產業別以科技業41%最高,其次是醫療業39%。

AI Agent自建軟體有哪些風險?

主要風險是後續維護:AI生成的程式碼一樣會累積技術債,而且沒有SaaS廠商提供的SLA、資安稽核支援與客服團隊。

企業該全面用AI Agent取代軟體採購嗎?

不建議。核心且需長期穩定支援的系統(如財務、HR)仍建議採購現成方案;內部效率工具、一次性流程才是AI Agent自建的甜蜜點。

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Google Ironwood TPU v7 推理專用晶片解析:效能追平 NVIDIA、成本低 44%,AI 晶片戰爭正式開打 | Google Ironwood TPU v7 Explained: Matching NVIDIA Performance at 44% Lower Cost — The AI Chip War Heats Up

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code