Arun Rajkumar

244 posts

Arun Rajkumar banner
Arun Rajkumar

Arun Rajkumar

@mickyarun

Co-Founder & CTO @PayWithAtoa | Building the future of UK payments | Open Banking | AI | FCA-Authorised | IIM Calcutta | We don't hire resumes. We hire intent.

Bengaluru Katılım Eylül 2008
232 Takip Edilen73 Takipçiler
Sabitlenmiş Tweet
Arun Rajkumar
Arun Rajkumar@mickyarun·
I've spent UK FinTech Week reading dev threads on open banking. Same misconceptions every time. PSD2 is "OAuth for banks." Payment Initiation is "just a bank transfer." Open banking is "Stripe but cheaper." None of that is right. Quick thread on what we got wrong (and learned). 🧵
English
1
0
1
86
Arun Rajkumar
Arun Rajkumar@mickyarun·
Friday afternoon. Half the eng teams I know are running sprint retros right now. Mine isn't. We stopped doing sprints months ago. Building what comes next. More soon. #BuildInPublic #AI #StartupLife
English
1
0
4
22
Jhon
Jhon@Jhoncls4·
@mickyarun Hi! Would you like to list your project on a Product Hunt alternative for free? They’re looking for unique projects, and yours would be a great fit.
English
1
0
0
7
Build in Public
Build in Public@buildinpublic·
What are you working on this week?
English
402
2
176
21.4K
Dr. DIG
Dr. DIG@CiprianiRanieri·
Founders only. Drop one number you're proud of. Followers. Users. MRR. Whatever. I'll go first. 1,150+ builders. ↓
Dr. DIG tweet media
English
12
2
10
616
Arun Rajkumar
Arun Rajkumar@mickyarun·
@dgwbirch Tough discussion, also tough to balance cost vs adaptation. Let’s see what’s turning in over the years.
English
0
0
0
5
Google for Startups
Google for Startups@GoogleStartups·
What’s the most manual part of your dev cycle?
English
6
1
9
1.2K
Arun Rajkumar
Arun Rajkumar@mickyarun·
@AravSrinivas Subagents inside an agent is where the actual product lives — single-model demos undersell this. What broke first when you scaled the orchestration?
English
0
0
0
42
Aravind Srinivas
Aravind Srinivas@AravSrinivas·
A Google Earth + Flight Simulator, fully cooked by Perplexity Computer with Codex/CC as subagents inside it. GTA vibes.
English
50
42
653
54.6K
Arun Rajkumar
Arun Rajkumar@mickyarun·
@sytaylor Onchain BaaS for agents is the bit — agents will hold balances long before humans trust them with cards. What stops a regulated bank from launching the same surface, just slower?
English
0
0
0
44
Simon Taylor
Simon Taylor@sytaylor·
🚨 JUST IN: Former PayPal CEO David Marcus just unveiled a stablecoin banking product for businesses and AI agents. Stablecoin balances. Yield. Payments. Cards. All chain-native, all behind an API. This is Banking-as-a-Service but onchain. --- Banking-as-a-Service is getting rebuilt on stablecoin rails. The old stack - Middleware (e.g. Unit / Synctera) - Galileo - FBO accounts at a sponsor bank There's now a chain-native version. - Squads Grid does it on Solana. - Bridge built it for fintechs, payroll and non banks - BVNK ships it embedded. Lightspark now joins them, with distribution into 65 countries and a Bitcoin L2 underneath. --- I keep thinking about how the "account" itself is being unbundled. What used to be a sub-ledger entry at a partner bank becomes a programmable smart account that holds dollars, settles 24/7, keeps its own yield, issues cards, and accepts agent-scoped permissions. The bank charter turns into a feature you plug in for accessing other rails. --- The agent piece is marketing, but its also more than that. Lightspark is already inside Google's AP2 protocol. As agents want to become platforms in their own right, re-sell financial services to their customers, they'd want a BaaS like thing. Why not go for something that uses stablecoins? --- Marcus has THE most fascinating back story - Former CEO of PayPal, has moved money traditionally - Behind Libra at Meta a global bank account and "stablecoin" that regulators pushed back on. Now this is a global stablecoin bank account distributed through an API, post-GENIUS Act, sold to businesses and machines. Sometimes timing is everthing. Does this compliment or commoditize BaaS over time? I expect the answer becomes obvious within 18 months.
Simon Taylor tweet media
English
10
12
153
18.8K
Arun Rajkumar
Arun Rajkumar@mickyarun·
@InnFin @JanineJoyHirt @neilshahuk @LSEGplc PISCES is the most underwatched UK structural reform of 2026 — pre-IPO liquidity is half the listing problem. Will it actually shift founder behaviour, or just ease the late-stage cap table?
English
0
0
2
41
Arun Rajkumar
Arun Rajkumar@mickyarun·
@sytaylor @mercury The teams that ship CLIs first usually ship better APIs second — the constraint forces clean grammar. Who's done it best so far in payments?
English
0
0
0
47
Arun Rajkumar
Arun Rajkumar@mickyarun·
@AlexH_Johnson CLI-first is also where agent-readability lives. Your stack is already agent-ready or it isn't — the bot just exposes the gap. What's the cleanest CLI you've seen out of fintech?
English
0
0
0
29
Arun Rajkumar
Arun Rajkumar@mickyarun·
@harshilmathur That phrase is a tell. The last 6 months changed what one engineer can ship in a day, not the model. What's the call you wish you'd made 6 months earlier?
English
0
0
0
62
Harshil Mathur
Harshil Mathur@harshilmathur·
If a leader starts with “AI isn’t new to us, we’ve been doing it for 4 years..." There’s a good chance they’ve missed what actually changed in last 6 months.
English
17
9
185
10.1K
Arun Rajkumar
Arun Rajkumar@mickyarun·
Confession: most monitoring stacks I've inherited at startups were over-engineered for the stage of the company. Tried Beszel this week as a counter-experiment. 23 MB hub. 6 MB per agent. SSH-based pull. Docker container stats automatic. Zero YAML. Zero PromQL. It gets you about 80% of what you actually open a dashboard to look at — in an afternoon. Not built for 1,000-server fleets. Perfect for the staging box, the side project, the small team that just wants to know if a container died. The AI tooling cycle has been loud for weeks. Nice reminder that not every productivity win needs a model behind it. github.com/henrygd/beszel #OpenSource #DevOps #SelfHosted
English
0
0
0
20
Arun Rajkumar
Arun Rajkumar@mickyarun·
A consultant told me last week she finishes the work in two days. Then spends two hours every Friday building invoices. PDF templates. Manual customer entry. Email it. Hope it doesn't land in spam. Wait three weeks. Chase. Repeat. We rebuilt invoices at @PayWithAtoa. Branded. Auto-filled from the last customer. Partial payments allowed. Paid straight from the customer's bank app. Half the cost of cards. FCA-authorised. Money in the account the same day. Stop building invoices. Send them. paywithatoa.co.uk #OpenBanking #UKFintech #Payments
Arun Rajkumar tweet media
English
0
0
0
20
Arun Rajkumar
Arun Rajkumar@mickyarun·
Anthropic just published a postmortem. Three back-to-back regressions in Claude Code in six weeks. Reasoning got dumbed down. A cache bug made it forgetful. A system prompt change tanked coding quality. The line every "fire your developers, AI does it now" founder should print and frame: "This change made it past multiple human and automated code reviews, as well as unit tests, end-to-end tests, automated verification, and dogfooding." Six layers of automated checking. Caught by humans reading feedback. Their own older model didn't find the bug. The newer one did — only after an engineer fed it the right repo context. I lead engineering at a payments platform. We use AI everywhere — code review, migration, scaffolding, evals. It's a force multiplier. Every guardrail it runs inside — the lints, the eval suites, the alert thresholds, the rollback plan — was designed by an engineer who understood the failure modes. AI is a force multiplier on a competent team. It is not a substitute for one. Tools don't ship software. Engineers do. anthropic.com/engineering/ap… #AICoding @claudeai #Engineering #BuildInPublic
English
0
0
0
80
Jonathan
Jonathan@joni_vrbt·
If you’re a vibe coder, let’s connect
English
168
1
216
10.4K
Audiencon⚡️
Audiencon⚡️@audiencon·
Which model are you using as your daily driver right now? > Codex > Claude > GPT-5.5 > Other
English
40
0
23
2K
Arun Rajkumar
Arun Rajkumar@mickyarun·
Here's what the single-shot benchmark missed: Codex won my engineering eval. Claude won my heart. Two more prompts + browser automation, and the site went live at brca.in — running Claude's design. Best instincts ≠ best collaborator. Different questions. 🔗 brca.in for more detail post dev.to/mickyarun/i-as…
English
1
0
0
32
Arun Rajkumar
Arun Rajkumar@mickyarun·
My son's cricket coach asked me for a website. Same prompt to 3 coding agents: Codex GPT-5.5 Claude Opus 4.7 Gemini 3.1 Pro Five total runs at different effort levels. The winner wasn't decided by the model. It was decided by taste. 🧵
English
1
0
0
79