AI Disrupting Software? What Singapore Businesses Should Do Next

US-listed software and services stocks have shed roughly US$830 billion in market value since late January, and the trigger wasn’t a weak earnings season or a macro shock. It was anxiety: investors are suddenly pricing in the idea that large language models (LLMs) can eat the “application layer”—the profitable software categories businesses pay for every month.

If you’re running a company in Singapore, this matters even if you don’t own a single tech stock. The selloff is a loud signal about where AI is heading next: beyond chatbots and “nice-to-have” productivity boosts, into core workflows in legal, finance, sales, marketing, analytics, and coding—the exact places most SMEs and mid-market firms spend on SaaS.

Here’s the stance I’ll take: AI isn’t going to wipe out software. But it will force a reset in how we choose tools, negotiate contracts, design processes, and prove ROI. Singapore businesses that treat this as a buying opportunity—not a panic moment—will be in a stronger position by end-2026.

Source context: Reuters reporting via CNA on the software selloff and investor debate about AI’s “existential threat,” including market moves and commentary from Nvidia’s CEO and major analysts. Landing page: https://www.channelnewsasia.com/business/software-selloff-continues-investors-debate-ais-existential-threat-5907141

The selloff isn’t about AI hype—it’s about AI getting practical

Answer first: Investors reacted because LLMs are moving from “assistive” features into end-to-end task execution, which threatens the pricing power of many software categories.

The CNA/Reuters piece points to a specific catalyst: a new legal tool built around Anthropic’s Claude, positioned as an agent-like plug-in that can operate across tasks like legal work, sales, marketing, and data analysis. Whether that single product wins or not, it represents a clear direction: LLM vendors want to sit between you and your existing software.

Why the market cares is straightforward:

• Software margins depend on habit and switching costs. If an AI agent can draft, research, classify, summarise, and route work across tools, users start asking why they’re paying for multiple subscriptions.
• The “application layer” is where recurring revenue lives. Infrastructure players sell compute; application vendors sell workflows. AI companies now want the workflows too.
• Business forecasting got harder. The article notes the standard 3–5 year outlook doesn’t fit an environment where capabilities jump every quarter.

For Singapore operators, the takeaway isn’t “cancel your SaaS.” It’s: your software stack is now a competitive decision, not an admin decision.

A useful mental model: “AI will unbundle, then rebundle”

Here’s what typically happens when a new platform wave arrives:

1. Unbundling: New entrants offer a cheaper or faster way to do a slice of the job (e.g., AI-assisted contract review, AI-generated prospecting emails).
2. Rebundling: Winners package those slices into a workflow suite (e.g., intake → analysis → approval → filing) and raise prices once embedded.

Most companies get this wrong by buying point solutions during unbundling—then discovering later they own six AI tools that don’t talk to each other.

“Is AI going to replace enterprise software?” Not the way people think

Answer first: AI will replace parts of software usage (especially routine tasks), but systems of record and compliance-heavy workflows won’t disappear—they’ll get an AI interface.

Nvidia CEO Jensen Huang reportedly called fears that AI would replace software “illogical.” I mostly agree with the spirit, even if the market is right to worry about disruption.

The reality: enterprise software isn’t just UI and features. It’s also:

• Data models and audit trails (who changed what, when, and why)
• Role-based access control and security policies
• Integrations with finance, HR, CRM, procurement, and customer support
• Regulatory expectations (especially relevant in Singapore for finance, healthcare, and regulated services)

LLMs are great at generating and transforming content. They’re not inherently designed to be your source of truth. That’s why, over the next 12–24 months, the winning pattern for most Singapore businesses will be:

AI layer (agents/copilots) + reliable systems of record (ERP/CRM/document management) + governed data access.

Where the real vulnerability sits: “workflow software” and “expertise software”

Some categories are more exposed than others:

• Routine workflow SaaS: ticket triage, email sequencing, meeting notes, simple dashboards
• Knowledge-heavy tools: legal research, compliance interpretation, policy drafting
• Coding and QA tooling: AI can automate boilerplate and testing faster than many teams expect

In the article, names like Thomson Reuters, RELX, MSCI, and exchanges/data businesses show up because they’re monetising information + workflow—a combination AI is now good at re-packaging.

What the software selloff teaches about choosing AI business tools in Singapore

Answer first: Don’t buy “AI features.” Buy measurable outcomes, solid governance, and portability so you’re not trapped when the market shifts.

This post sits inside our AI Business Tools Singapore series, so let’s make this practical. When investors say “moats look narrower,” businesses should translate that into procurement rules.

1) Demand ROI you can audit (not vibes)

If your vendor can’t help you quantify ROI, you’ll struggle to defend budgets when leadership asks “why are we paying for this?”

A simple ROI template that works:

• Time saved per role per week (e.g., 2.5 hours)
• Fully loaded hourly cost (e.g., S$55/hour)
• Adoption rate (e.g., 60% of team)
• Error reduction / rework reduction (e.g., -20% revisions)

Then calculate: (time saved × hourly cost × adoption) − subscription cost.

Be strict: if the tool doesn’t clear a hurdle rate in 60–90 days, it’s a pilot, not a rollout.

2) Prioritise data governance like it’s a product feature

For Singapore firms handling customer data, contracts, or regulated documentation, governance is non-negotiable.

Checklist for AI tools:

• Can you control what data is sent to the model?
• Is there tenant-level isolation and clear retention policy?
• Can you implement role-based access and approval flows?
• Do you get logs for prompts, actions, and outputs?

If the answer is fuzzy, pass. Cheap tools get expensive when you’re cleaning up incidents.

3) Avoid vendor lock-in by designing for portability

The selloff highlighted volatility. Your tech stack should assume volatility too.

Design choices that keep you flexible:

• Keep core data in systems you control (CRM/ERP/data warehouse)
• Use tools that support standard exports and APIs
• Document prompts/workflows in a shared internal library
• Don’t build your business logic only inside one vendor’s “magic” automation

This is future-proofing in a very boring sense—and boring is good.

Practical playbook: adopt AI without getting disrupted by it

Answer first: The safest way to adopt AI is to start with high-frequency workflows, build a governed “AI operating layer,” and keep humans responsible for final decisions.

Here’s a playbook I’ve seen work for SMEs and mid-sized teams (and it maps to how AI is encroaching on applications):

Step 1: Pick 3 workflows where speed matters and mistakes are tolerable

Good candidates:

• Customer support: summarising tickets, drafting responses, tagging and routing
• Sales: account research, call summaries, follow-up drafts
• Marketing: first drafts, ad variations, content repurposing

Avoid starting with high-risk workflows like legal sign-off or financial reporting until governance is mature.

Step 2: Define “human-in-the-loop” rules

Write it down. Examples:

• AI can draft; humans approve anything customer-facing
• AI can suggest actions; humans execute actions with financial impact
• AI can summarise; humans verify quotes, numbers, and commitments

This stops the common failure mode where teams treat AI output as authoritative because it sounds confident.

Step 3: Build a small internal “prompt + policy library”

A simple Notion/Confluence page is enough:

• Approved prompts per function (sales, ops, HR)
• Brand voice guidelines for generated content
• Do-not-use rules (confidential data, personal data)
• Examples of “good output” vs “unacceptable output”

The fastest way to scale adoption is to remove guesswork.

Step 4: Measure adoption like you measure revenue

Track:

• Weekly active users
• Number of tasks completed with AI assistance
• Average time saved per task
• Escalations/rework rate

If adoption stalls, the tool is either poorly integrated into workflows or not delivering value. Don’t blame “change resistance” until you’ve fixed friction.

“People also ask” (and the real answers)

Will AI reduce my SaaS costs in 2026?

Yes, if you rationalise your stack. Many companies can cut tools that primarily handle drafting, summarising, tagging, and basic analytics—if they have a single AI layer that does those tasks across systems.

Should I wait until the market settles?

No. The market may stay volatile, but operational learning compounds. Run small pilots now so you know where AI improves throughput and where it creates risk.

What’s the biggest risk for Singapore SMEs adopting AI?

Shadow AI. Teams will use consumer tools with company data if you don’t provide approved options. That’s a governance problem, not a people problem.

What to do this quarter if you’re serious about AI business tools

The software selloff is a financial story, but the operational message is simple: AI is competing for your workflow budget. Use that pressure to get disciplined.

Start with these actions in the next 30 days:

1. Inventory your top 15 software subscriptions and label each as: system of record, workflow tool, or nice-to-have.
2. Select one approved AI workspace (with governance) for company use.
3. Pilot 2–3 workflows with clear ROI targets and human-in-the-loop rules.
4. Renegotiate renewals: push for shorter terms, usage-based pricing, and exit clauses.

If you’re following our AI Business Tools Singapore series, you’ll recognise the throughline: tools don’t create advantage by existing in your stack. Advantage comes from process design + governance + measurable outcomes.

AI isn’t an existential threat to your business—confusion is. The companies that win in 2026 will be the ones that can answer a basic question quickly: Which parts of our work should be automated, which parts should be augmented, and which parts must remain human-owned?