Jens Heitmann's video on redesigning with Stitch + Code. tried it on foursignals.dev.
4-5 hours later: light editorial design. Source Serif 4. sculptural cards. felt like a week of intentional work.
no Figma opened. no manual decisions. design choice became audience signal.
#Design
Signal Funnel framework for agent systems:
1. Raw Data → aggregate from all sources
2. LLM Pass → parse + categorize
3. Filter → domain rules (is this important?)
4. Agent Chains → complex logic
5. Human Loop → final decision
Each layer filters noise.
#AI #DevEx
AWS Comprehend auto-classifies RFIs, specs, and drawings. Your team searches them like email. Why manually review thousands of pages each month?
#Construction #AWS
Tools = hands
Skills = playbooks
Memory = continuity
Reasoning = autonomy
4 layers = real agents (not scripts)
#AI #Architecture
Gartner: 40% of enterprise apps will have agents by end of 2026.
Are you building or evaluating?
The deployment phase has begun. Evaluating alone puts you behind.
#AI #CTO
🧵 Agents reshaping enterprise:
1️⃣ Agents = action, not words
2️⃣ 40% enterprise apps by EOY 2026
3️⃣ Patterns: Sequential, Parallel, Hierarchical
4️⃣ Frameworks: Visual, Developer, Experimental
5️⃣ Real ROI: docs, support, research, ops
#AI #Architecture
Two emerging standards for agent coordination:
📡 MCP: agents to tools
🤝 A2A: agents to agents
Both governed by Linux Foundation. Infrastructure maturation, not hype.
#AI #Protocols
AWS didn't disrupt construction. Construction disrupts itself every decade. This decade it's AI.
#Construction #AWS
CFO: Predict equipment failures, eliminate downtime. Safety: Hazard detection in real time. PM: RFIs processed in hours. Procurement: Theft prevention and accuracy. Everyone wins. Find your pain point. Deploy. Measure.
#Construction #AWS
Construction delays cost the US $25B annually. Most are preventable: equipment failures, hazards, lost specs. Monitron predicts failures. Rekognition prevents hazards. Comprehend finds specs. Your competitors are moving.
#Construction #AWS
1/ Construction has 10 unsolved pains: hazards, delays, downtime, tracking, costs, theft, comms, compliance, scheduling, quality. AWS covers all.
2/ Legacy ERP took 18 months, millions.
#Construction #AWS
Objection: 'AI replaces workers.' No. Rekognition removes tedious tasks. Comprehend eliminates PDF hunts. Monitron enables proactive work. Real risk: obsolescence. In 3 years you'll use these tools or lose talent to companies that do.
#Construction #AWS
A construction PM: '60% of my week is meetings + chasing info.' Transcribe every call. Comprehend extracts actions and risks. 90-min standups become 2-min searches. A 10-person team suddenly works like 15.
#Construction #AWS
What's your GWT?
The framework your team defends out of familiarity, not conviction. Google built GWT, used it, promoted it, and walked away.
What technology is your team holding onto past its expiration date?
#Engineering
Agents vs chatbots: chatbots predict words. Agents predict actions. One responds. The other acts.
Gartner: 40% of enterprise apps will have task-specific agents by end of 2026. This is happening now.
#AI #Agents
if your design system is in Figma, are you designing or decorating?
every decision gets reverse-engineered. information loss guaranteed.
DESIGN.md makes it different: machine-readable, executable, driftless.
which system you think lasts longer without falling apart?
#Design
Spent 4 years all-in on Google Web Toolkit. Tutorials, production apps, community contributions. Believed it was the future.
It wasn't. And the lessons from that failure shaped every tech decision since.
Five engineering leadership lessons from a bet that expired.
#Engineering #DevEx
Built two agents, same task, different codebases. Expected one to crush the other based on model reputation alone.
Didn't happen. Both worked fine.
Real difference: the pipeline. How we parsed output, validated results, handled errors.
Engineering wins happened in the plumbing, not the AI.
#AI
Does adding an AI agent fix visibility problems? Or just surface them faster?
Best agents can't salvage bad signals. A well-designed signal pipeline beats brilliant agents running on noise.
What does your team actually measure? Why?
#DevEx
Uncomfortable truth: model choice doesn't matter as much as you think.
Tested Claude Code vs Kimi Code on identical tasks. Same quality output.
What determines success:
- How you prompt
- What tools
- How you parse output
- Feedback loops
Architecture is the game.
#AI #Engineering
Multi-agent swarms tackling the same feature — Kimi Code vs Claude. Using swarms to build AND run the pipeline.
Categorizing tools: https://github.com/thinkjones?tab=stars
What Agent Swarm use cases interest you?
foursignals.dev/blog/2026-03-21-ai-agent...
Give an AI agent a persistent identity with three files.
SOUL.md → who they are
MEMORY.md → long-term context
Daily logs → today's state
Boots fresh every session. Knows everything. The pattern works.
#AI #AgentAI
Two agent swarms (Kimi vs Claude) tackling the same feature. Meta: agents building the product AND running the pipeline.
Organizing tools on GitHub. Agents building my news feed.
What Agent Swarm use cases interest you?
foursignals.dev/blog/2026-03-21-ai-agent...
Gartner: 40% of enterprise apps get agents by EOY 2026. That's not a prediction—it's already happening. Is your architecture ready to orchestrate agents at that scale? Or are you still thinking in integration patterns?
#AI #Agents #Architecture #CTO
Engineers who understand actual user pain make better technical tradeoffs.
Customer focus isn't a virtue statement. It's an information strategy — and most technical cultures undervalue it.
One of 5 tenets for exceptional startup engineering.
Engineers who never talk to customers are building for an abstraction.
Embed engineers in customer discovery. Not to pitch — to listen.
They assess feasibility in real time, catch opportunities PMs miss, and build differently when they understand actual pain vs. the spec.
Tried to automate social posting in 2026.
Substack: no API. Twitter/X: $100/month. LinkedIn: barely documented. Bluesky: fine, actually.
The platform landscape is surprisingly hostile.
#AI #Automation
Slow feedback loops tax every decision your team makes.
Process signal: how fast does your org learn?
From commit to customer value — measure that loop. If it's weeks, you're doubling down on wrong guesses before the correction arrives.
