DataPA

Combining a wealth of experience delivering analytics to industry, with a drive to innovate, we help organisations understand their business.

[inmydata @inmydata]

Your AI can architect software. It can't reliably add up a column of numbers.

Both true, same model, at the same time. That gap is the most important thing to understand about building on these systems.

So we stopped asking it to add up https://t.co/NreZ3SyYQJ

[inmydata @inmydata]

In seven days, GitHub repriced annual Copilot and Anthropic A/B tested Claude Code out of the $20 plan.

Flat-rate billing has broken under #agentic load.

The fix is in the architecture, not the invoice.

https://t.co/dnYbLFWgY2

[inmydata @inmydata]

Great to see Scotland's tech community gathering again at #ScotSoft2026.

Our CTO Nick Finch will be on stage exploring how agentic AI is reshaping software UI, and what that means for the next wave of products built here. See you at the EICC in September.

[inmydata @inmydata]

@SpaceX bought @Cursor. @Microsoft owns @Copilot.
@Anthropic owns @ClaudeCode. @OpenAI owns @Codex.

Your coding tool is a supplier now, not a productivity choice.

How we build so we're not hostage to any of them https://t.co/g0iMzIRhIX

[inmydata @inmydata]

#Opus 4.7 deprecated temperature, top_p, top_k, and extended thinking budgets

In their place, one new parameter. task_budget

Stop tuning. Start bounding how much thinking the model can afford

For some, a config change, for others a rewrite.

https://t.co/bzYo34uggt

[inmydata @inmydata]

Organisations struggling with #AI costs are building #agents like #chatbots. Fire a prompt, get a response, worry about the bill later. That doesn't work for #agentic systems. Constraint isn't the opposite of capability. It's what makes capability viable https://t.co/lcn7XcAVOT

[inmydata @inmydata]

Computing solved the structured data problem fifty years ago. The other eighty percent, the expertise and judgment that actually drives decisions, has never been computerised. Until now. https://t.co/xkagUR0SNh #AI #RAG

[inmydata @inmydata]

We gave #AI's most valuable capability a name that implies it's broken. #Hallucination and creativity are the same mechanism. The variable is constraint. https://t.co/BFlxs3GQbL

[inmydata @inmydata]

Every major vendor just shipped agent #identity products. None of them can make you trust your agents.

Coding #agents solved this months ago. #Trust escalation. Start constrained, earn #autonomy, audit everything.
https://t.co/QDGhkfWGWf

[inmydata @inmydata]

Chroma Research tested 18 frontier models. Every single one gets worse as you fill the context window.

The bottleneck wasn't capacity. It was curation.

Read what #ContextEngineering actually looks like when you're building it, not just talking about it https://t.co/zB8RdcvX7m

[inmydata @inmydata]

Three #MCP servers in production. One serves operational #data. One serves expert knowledge via #agentic #RAG. The third lets agents request new knowledge by interviewing experts.

Three types of intelligence, one protocol. Composability is the unlock. https://t.co/HkhIysQ1p6

[inmydata @inmydata]

We built two AI agent systems in 4 months. Work that would have taken 2 years.

The secret isn't the model. It's that AI built the systems too. Data pipelines, business logic, retrieval architecture.

Be AI-first in how you build, not just what you build.

https://t.co/JeMG0OMJEF

[inmydata @inmydata]

The narrative that agentic coding = technical debt is not wrong.

It's just aimed at the wrong people.

Vibe coding and agentic engineering are not the same thing. Discipline is the variable.

https://t.co/GlCrKiQqPA

#Agentic #SoftwareDevelopment #AI #VibeCoding

[inmydata @inmydata]

The incumbents still here in 2030 won't be the ones that bolted #AI on and called it an #agent

They'll be the ones who accepted the seat model is broken and competed on what you can't recreate over a weekend

30 years of the right data. That's the moat
https://t.co/7vZzSm7jRs

[inmydata @inmydata]

1:30pm - started building a pen testing suite
3:30pm - finished
4:00pm - first scans running
5:00pm - issues fixed, gone to the pub

A commercial pen test costs £5-25k and takes weeks.

Read why this pattern explains the turbulence in software stocks https://t.co/oZpcEeTmw5

[inmydata @inmydata]

On February 3rd, $285 billion was wiped from #software stocks in a single day. By mid-February, a trillion.

It wasn't a #bubble bursting. It was #AI getting good enough to do the work those companies charge for.

That money didn't disappear. It moved https://t.co/oZpcEeTmw5

[inmydata @inmydata]

There is no #AIBubble. But software is being disrupted.

#AI already delivers value. The divide is between those architecting for it and those dismissing it as hype.

Only one group stays competitive

https://t.co/17xvpXuOEM

#AgenticAI #SoftwareIndustry #EnterpriseAI #inmydata

[inmydata @inmydata]

#AI #slop isn’t a model problem. It’s a process failure.

Treat #LLMs like vending machines and you’ll get average output.
No plan. No review. No judgement.
AI can generate prose.
Only process creates meaning.

Slop is what happens when you skip the work.

https://t.co/SFvLKiTKvL

[inmydata @inmydata]

We finally got the AI assistant people wanted.
Not because it answers better, but because it gets things done. That’s also why it’s dangerous. Alexa is safe because it can’t do much. @openclaw is exciting because it can do almost anything. https://t.co/bi4cF4Aid2 #AgenticAI #AI

[inmydata @inmydata]

Software spent 40 years piling up menus, forms and workflows so humans could translate intent into clicks.

Agents invert that. You state the outcome. They do the clicks.

The UI becomes a readout not the control panel. https://t.co/pUgkqJRufl