Singapore's three workhorse grants close in 20 days. The replacement still has no numbers.
20 days to 29 September.
PSG, EDG and MRA — the schemes most Singapore SMEs have used to part-fund system work, from ERP and CRM upgrades to integration and automation — cease on 29 September 2026 and consolidate into a single scheme, EDGE. EnterpriseSG has confirmed the sunset but not yet published EDGE's support levels, annual cap, or opening date beyond "H2 2026." A known grant closing beats an unknown grant opening.
You're an SME with a project already scoped — get the application in before 29 September under PSG's known terms (up to 50% support, S$30,000 cap) rather than wait on a scheme whose numbers don't exist yet.
You're an SME without a project yet — don't invent one to catch the deadline. Projects reverse-engineered from a grant are the ones that stall six months later.
You're not an SME — PSG excluded you entirely. EDGE is expected to widen eligibility to non-SME businesses, but there's nothing to file yet — worth watching for the opening date rather than acting now.
Does this affect you? The week's announcements, translated
You see a dozen AI launches a week. Most change nothing about how your business runs. Here are the three from this week that might, and what each is actually worth to you.
A global survey just confirmed what the skeptics suspected about AI ROI
What happened. McKinsey's annual State of AI survey (1,719 respondents, 97 countries, fielded 4 May–8 June, published 25 August) found regular AI use in at least one business function at nearly nine in ten organisations, and enterprise-wide scaling at 44%, up from 38%. The number that didn't move: only 37% attribute any EBIT impact to AI, flat on 2025, and just 6% qualify as high performers capturing 5%+. What separates the two groups: 73% of high performers redesigned workflows around AI, versus 25% of everyone else.
Matters to you if you're fielding "why isn't our AI spend paying off" from a CFO or board. The 73/25 split is the answer — budget spent without process redesign produces individual productivity anecdotes, not P&L movement.
You can ignore it if you're already leading every AI conversation with workflow redesign rather than tool selection — this just hands you the number to back it up.
Vendors are testing pay-per-outcome pricing. Don't sign it yet.
What happened. Zendesk, Pegasystems and HP are each piloting a shift from per-seat billing to billing on completed work — a fee per AI-resolved ticket, per completed case, or a ticket-reduction guarantee. Gartner's Tom Coshow supplies the reality check: 19% of services buyers use outcome pricing today, and he projects under 25% of tech CEO services contracts will by 2031 — "more buzz than reality" (CIO Dive, 31 August).
Matters to you if a vendor pitches outcome-based pricing this quarter. Treat it as a negotiating opening, not an adopted standard, and insist the billable "outcome" is defined in the contract before you sign — an undefined "resolution" or "completed case" is the dispute that surfaces at renewal.
You can ignore it if every contract on your desk is still per-seat and no vendor has raised this with you yet.
A 75% price cut just changed the math on every agent proposal you've costed
What happened. Anthropic released Fable 5.1 and Mythos 5.1 on 3 September with a 75% cut to cache-read pricing. Cache reads dominate the running cost of any long-lived agent that carries a large stable context — a codebase, a policy library, a case file — so this doesn't add a capability, it changes the arithmetic on workloads that looked marginal a month ago.
Matters to you if you costed an agent proposal on pre-3-September inference prices — re-run the unit economics before you present it. The same caveat applies to costing on the assumption prices keep falling: fine to use current prices, risky to bank on the next cut.
You can ignore it if nothing you're evaluating is a long-context, long-running agent.
What going first exposed: banks running AI they can't stop
Wolters Kluwer's US Banking AI Risk and Governance Index for H1 2026 (230 US banking professionals, published 10 June) found 34% of banks have no documented AI kill-switch protocol, and 38% can't report an AI failure to regulators when one happens. Asked where agentic AI carries the most risk from insufficient human oversight, 33% named lending and underwriting, 30% named collections.
The gap isn't model quality — it's incident machinery: a documented stop condition, a named owner, a reporting path. Roughly a two-week engagement, board-legible, and in a regulated sector it's what actually unblocks deployment rather than decorating it. US data, but a MAS-supervised institution faces the identical question.
The governance headline everyone misread this week
The Digital Omnibus deal, reached politically 7 May 2026, pushed the EU AI Act's standalone high-risk system obligations to 2 December 2027 and product-embedded high-risk systems to 2 August 2028, grandfathering systems already on the market unless substantially modified. That's been widely read as "the AI Act is postponed." It isn't, entirely: Article 50 transparency obligations — disclosure for chatbots, emotion recognition and deepfakes — took effect 2 August 2026, and the watermarking grace period for generative systems already on the market expires 2 December 2026.
Any client-facing chatbot serving EU users is in scope now, and watermarking grace runs out in under three months. Cheap to fix today, embarrassing to fix in December.
Government initiatives
What Singapore is funding and building this quarter, as distinct from what it's regulating.
PSG / EDG / MRA close 29 September 2026, consolidating into EDGE, expected to widen eligibility to non-SMEs. EnterpriseSG hasn't yet published EDGE's support levels, caps, or launch date beyond H2 2026 — worth contacting every prospect with an active PSG-eligible project this month.
National AI Impact Programme (announced 2 March) targets 10,000 enterprises and 100,000 "AI Bilingual" workers over three years, raising the AI-enabled share of grant-supported solutions from 30% to 50%. The route in is IMDA pre-approval, which requires a market-proven solution from a reputable vendor — productised qualifies, bespoke consulting doesn't.
Digital Leaders Accelerator Bootcamp, delivered with EY-Parthenon and BCG, targets 2,000 Digital Leaders over three years; IMDA says it will "progressively onboard new industry partners" — an open door to a panel currently staffed by two global firms.
Hiring signals
US and Singapore data on what's happening to AI-related roles.
Job postings requiring AI skills rose 27% between April and August 2026, and 165% year on year (Lightcast data) — growth is accelerating, not levelling.
The fastest-growing complementary skills in those postings are automation, workflow management and operations, not modelling — the same finding as the McKinsey 73/25 split above, arriving through a different door.
UBS has made AI proficiency a formal requirement for its 2027 graduate and intern investment banking class (FT, 6–7 Sep). Expect it to become a screening criterion rather than a specialism across professional services.
McKinsey found 39% of respondents expect AI-driven job cuts, up from 32% — but actual 2025 reductions fell well short of the prior year's predictions. Anticipation is running ahead of action; treat client headcount projections accordingly.