Inspiring to see deep-tech innovation and sustainability take center stage at the 18th #BRICSSummit.
ββ From energy-efficient software architectures to eco-centric engineering, technology must serve both business growth and the planet. ππ±
β#BRICS2026
The IAF/ILAC β Global ACI mark transition is paused until at least October. The CBs that will handle it cleanly treat the accreditation mark as data in the certificate register, not artwork on a template. Trigger day: a config change, not a fire drill.
From 11 September, an actively exploited vulnerability in a product sold into the EU is a 24-hour reporting obligation. Finance went through this. Carbon is going through it. Security assertion is free β evidence now has a deadline.
94% of tech leaders rate AI code higher quality at review. 78% see more production incidents after deployment. Review inspects an artifact; production interrogates behavior. When plausible code is free, gates that measure plausibility are worthless.
Financial statements crossed this line a century ago: no audit, no credibility. Sustainability reports cross it this December (ISSA 5000). AI systems are next. Every market ends up pricing evidence over assertion β and the evidence chain becomes the product.
Everyone in AI compute asks where to get more power. Almost nobody asks how to get more out of what's already running. Prove it at 2 MW first, emissions, water, heat reused in agriculture, registry-grade evidence, then extrapolate to larger data centres if the numbers hold.
The EUβs first permanent-removal methodologies (DACCS, BioCCS, biochar) are certification machinery: recognised schemes, accredited certification bodies, registries, reversal liability. A tonne is now a data object with an audit trail β or it isnβt a tonne.
Since January the two bodies behind global accreditation β IAF and ILAC β operate as one, with a single recognition arrangement across 140+ countries. The rail that carries trust just got simpler, right as the traffic on it β AI, carbon, ESG β multiplies.
Certification pricing has one variable: audit time.
IAF MD 5 mandays are the price of trust β and every undocumented reduction is a nonconformity waiting for a witnessed audit.
Manday justification is an auditable artifact, not tribal knowledge.
AI vendor trust is moving from assertion to evidence.
The questionnaire era accepted policy PDFs. The ISO 42001 era asks who verified that your controls operate β and it's entering RFP language already.
Verification is becoming the product.
High-risk AI obligations: deferred to Dec 2027. Accreditation for ISO 42001 scope β competence files, witnessed audits, impartiality review β takes most of that window. For certification bodies the deferral isn't a pause. It's the runway.
The EU moved its AI Act deadlines to 2027β28 and kept what matters: since August, AI systems must disclose themselves and synthetic media must carry machine-readable marks. Deadlines slip. The shift from assertion to evidence doesn't.
Half of fintech runs on Scala/Akka systems whose builders stopped applying to job posts. If your platform depends on that stack, you don't need a req to close β you need an engineering partner who lives in it. As an Akka partner, that's exactly what we built.
Global ACI transition, translated for CB owners: new mark on every new certificate (deadline is real), MD-series harmonisation by Dec, and one evidence trail your assessor can walk. Spreadsheets were fine in 2024. They are not a transition plan.
The AWS bill has three predictable leaks: on-demand compute running 24/7, staging sized like prod, observability growing faster than infra. Two weeks of senior DevOps attention fixes all three. You don't need a FinOps org. You need someone who's done it before.
The part of hiring Scala engineers that always surprises me isn't skill assessment. It's the timeline. You aren't just looking for Scala devs; you're looking for a small subset of the already scarce senior JVM talent pool. It always takes longer than I budget for.
@karpathy Spot on, @karpathy . The stamina is there, but you nailed the bottleneck: orchestration & closed feedback loops. As @elonmusk proved with FSD, AI needs real-time native perception to fix jank. Turning it into "ephemeral GTA" require Agentic AI pipelines we build at @fusionpact .
My proceeds from PayPal were $180m.
I put $100m in @SpaceX, $70m in @Tesla and $10m in Solar City.
I had to borrow money for rent.
βElon Musk (photo from 2008 after the third consecutive failure of his Falcon 1 rocket)